	          '  N     8 @p ~ n                   /hY                <     -,-p     ,-I-     I--    -4-    4-uC-    uC-v-t    v-30-     30-7-X    7--    -ub-p	    ub-߂-x
    ߂--
    -U-    U-ճ-\    ճ-׳-<                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     t            !          N!  1          1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        7  L          n                                                6  :          Y  ]	          |	  
          
                                              9         :  W                      ,         -  ?         @  R         S  e                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h         T$  Y%          x%  }&         vL  M         M  N         N  O         O  Q         Q  )R         *R  MS         NS  qT         rT  U         U  V         V  W         W  Y         Y  %Z         &Z  I[         J[  m\         n\  ]         ]  ^         ^  _         _  `         `  !b         "b  Ec         Fc  id         jd  e         e  f         f  g         g  h         h  j         j  Ak         Bk  el         fl  m         m  n         n  o         o  p         p  r         r  =s         >s  at         bt  u         u  v         v  w         w  x         x  z         z  9{         :{  ]|         ^|  }         }  ~         ~                                   5         6  Y         Z  }         ~             Ň         Ƈ                        1         2  U         V  y         z                                              
            .  3         R  t         v                                              (         *  L         N  p         r                        ܠ         ޠ              $         &  H         J  l         n                        ب         ڨ                       "  D         F  h         j                        ԰                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               O!         P!  1         1  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    Z
         [
             "                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    Z
         [
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-a-     a--      -1<-     1<-J-     J-~-     ~-]3-"     ]3--     -b-*     b-	n-R     	n--B     --                    =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     {            '!          X!  1          2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z         9   z   =   *   =   z   A   2   A   z   E   2   E   b   I      reason  OperatorKernel  value   3       kernelSpec      outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                 \               "              j             
            R0            G            ^            v            Z                 	            
       7            2N            be            |                                    "            R                                              6                       ,            C            "[            
s            :            "            
            z            b                 !       B    "       9    #       *S    $       n    %       ҅    &           '       2    (       b    )           *           +           ,       "(    -       R?    .       V    /       m    0            1       2,     2       bC     3       Z     4       q     5            6            7       "     8       "     9       "     :       z     ;       z    <       0    =       BJ    >       rP    ?       i    @       p    A       Z    B           C       *    D           E           F       *
    G       r!    H       8    I       O    J       g    K       2~    L       b    M           N           O           P       "    Q       R	    R            S       7    T       N    U       f    V       B}    W       r    X           Y           Z           [           \           ]           ^       Z9    _       P    `       bj    a           b       
    c       :    d           e            f            g       4     h       K     i       b     j       z     k       B     l       r     m            n            o            p       2    q       b    r       3    s       J    t       a    u       "y    v       "    w       R    x       R    y       :    z           {       z    |       "    }       B<    ~       rB           [           b           Z~                      *                                 r                       23           bJ           a           x                      "           R                                                       B2           rI           `           w                      2           b           b                                            J9           U           p           2                      ڹ           R                                            2           BI           r`           w           Ҏ                      2           b                                            "1           RH           _           v                                                                  2	           "           (           BB           rH           c           Z}                      B                                 J           	           .	           "F	           R]	           "t	           R	           	           	           J            z'            >            U            
m            :            j                                    :            :                       *           
G           ba           }                      J                                 j                      +           B           *Z           Zq                                                       J           z                      *           
B           :Y           jp                                 ʵ                                 "                      */           ZF           L           c           i          J                                                  J                    
;    	      :R    
      ji                    ʗ                    *          Z                              "          9          P          g                              J                              *          *          *          D          Ba           {    !      ʒ    "      B    #      r    $          %          &      J    '      z&    (      =    )      T    *      
l    +      :    ,      j    -          .          /          0      *    1      Z	    2      %	    3      <	    4      S	    5      j	    6      	    7      	    8      	    9      J	    :      	    ;      
    <      
    =      2
    >      8
    ?      P
    @      JV
    A      q
    B      
    C      z
    D      
    E      2
    F      
    G          H      j'    I      >    J      U    K      l    L      *    M      Z    N          O          P          Q          R      J    S      z&    T      =    U      T    V      k    W      ڂ    X      
    Y      z    Z      z    [          \          ]      b    ^      /    _      JK    `      d    a      {    b      b    c          d      
    e      :    f      j    g          h      &    i      =    j      *U    k      Zl    l          m          n          o          p      J    q      z    r          s      %    t      
=    u      zV    v      m    w          x      J    y          z          {      z    |      	    }      !    ~      J'          z>          D          
`          zz          ڕ          J                    
          R                    ,          *D          Z[          r                                        J          z                              
,          
C          :Z          :q          
          b          b                              2                    ;          T          k          j                      9           *Q           Zh                                                       J           z           
          !          8          O          f          
~          
          :                              J          z          )          
A          j]          t          ʋ                    *          Z                    B                    *          r0          K          b          J~          z                              
          :          j	                     7          N          *f          Z}                                                            B          r"          <          S          ro                    Z          ʽ                    Z                    $          <          JS          zj                    ژ          
          :          j                              #          *;          *R          Zi          Z                              *                              B          r+          F          ^          Ju          2          b                     Z                    B	          %	          =	          rY	          p	          	    	      J	    
      z	          	          	          

          :
          j/
          F
          ]
          t
          *
          Z
          Z
          
          
          
                    B1          K          b          B~          r                         !      J    "          #          $      :/    %      jF    &      ]    '      t    (          )      *    *      Z    +          ,          -          .          /      J.    0      zE    1      z\    2      s    3          4      ڡ    5      2    6      b    7          8          9      b     :      7    ;      
S    <      :j    =      j    >      R    ?          @      j    A          B      :    C          D      
*    E      RA    F      ]    G      t    H      Z    I          J          K          L          M      J    N      z    O      3    P      J    Q      
b    R      :y    S      j    T          U          V          W          X          Y      j    Z      4    [      
N    \      :e    ]          ^          _      Z    `          a          b          c      r    d      
(    e      :?    f      jV    g      m    h      ʄ    i          j      *    k      Z    l          m          n          o      '    p      J>    q      JU    r      zl    s      z    t          u          v      2    w          x          y      2    z      b0    {      K    |      
c    }      :z    ~      "          R          :                                        Z          5          R          Ji                    ڜ          
          :          j                              '          *?          ZV          m                              ҳ                                                  *          *D          Z[          v                    z                                                            
6          zO          f                    J          z                              
          :          j"          9          P          g          ~                    R                              
                    +          *G          Z^          u                                                  J                               "/          "G          j^          v          J          z                              
          :          j          /          F          ]          *u          Z          B          r          Z                              *          4          K          b          *z          Z                                                                      Z7          N          g          *                    ұ          J          z                              
*          
A          :X          :o          j          ڟ          
          z                    "           R           5           L           *d           *{                                  B                                 !          b!          b5!          L!          d!    	      {!    
      J!          z!          !          !          J!          z"          ""          9"          
Q"          :h"          j"          R"          "          "          J"          z"          #          "#          
:#          :Q#          jh#          #           ʖ#    !      #    "      #    #      *#    $      *#    %      Z
$    &      #$    '      :$    (      jT$    )      k$    *      $    +      B$    ,      $    -      $    .      $    /      J$    0      %    1      .%    2      2E%    3      ]%    4      Jt%    5      %    6      %    7      Z%    8      %    9      %    :      2&    ;      #&    <      :&    =      
R&    >      
i&    ?      :&    @      :&    A      &    B      &    C      &    D      Z&    E        	   F        	   G      7  	   H      O  	   I      Bf  	   J        	   K        	   L      R  	   M        	   N        	   O      *  	   P       	   Q      , 	   R      D 	   S      2[ 	   T      br 	   U       	   V       	   W      R 	   X       	   Y       	   Z       	   [       	   \      B6 	   ]      rM 	   ^      d 	   _      { 	   `       	   a      2 	   b      2 	   c      b 	   d      b 	   e       	   f        	   g      8 	   h      rR 	   i      i 	   j       	   k      J 	   l      · 	   m       	   n      " 	   o      R 	   p       	   q      j, 	   r      C 	   s      [ 	   t      r 	   u      " 	   v      R 	   w      ¼ 	   x       	   y      " 	   z      R 	   {       	   |      0 	   }      G 	   ~      ^ 	         v 	          	         Z 	          	          	         2 	         z 	          	         / 	         F 	         "^ 	         w 	          	         2 	         b 	          	         
 	          	         & 	         "> 	         W 	         bn 	          	          	         R 	          	          	          	          	         B2 	         rI 	         ` 	         w 	          	         2 	         2 	         b 	         b 	         	 	         	 	         4	 	         rN	 	         e	 	         	 	         J	 	         ³	 	         	 	         "	 	         R	 	         
 	         j(
 	         ?
 	         2W
 	         bn
 	         
 	         
 	         
 	         "
 	         R
 	         
 	          	         ' 	         ? 	         V 	         Bm 	         B 	          	         Ҳ 	          	         b 	          	          	         
' 	         "> 	         RU 	         o 	         z 	         Ҡ 	         ҷ 	         J 	         J 	          	         2 	         b6 	         O 	         g 	         b 	          	          	         " 	         R 	          	          	         * 	         B 	         BY 	         rp 	          	         Ҟ 	         ҵ 	          	          	         2 	          	         , 	         G 	         B^ 	         y 	          	         b 	          	          	          	         "	 	         % 	         < 	         Y 	         Bp 	         r 	          	         ҵ 	          	         2 	         b 	          	         b) 	         @ 	         W 	         n 	           	         
 	         R 	          	          	         * 	         B 	         ( 	         ? 	   	      V 	   
      *q 	         Z 	          	          	         * 	         * 	          	         ! 	         B8 	         Q 	         h 	         Z 	          	          	         2 	         b 	          	          	         * 	         "B 	         RY 	         p 	           	   !       	   "       	   #       	   $       	   %      B 	   &       	   '      + 	   (      RE 	   )      \ 	   *      w 	   +      * 	   ,       	   -       	   .      B 	   /       	   0       	   1      B( 	   2      r? 	   3      [ 	   4      s 	   5      2 	   6      b 	   7       	   8       	   9       	   :      " 	   ;      R 	   <      , 	   =      C 	   >      Z 	   ?      r 	   @      B 	   A       	   B      ҷ 	   C       	   D      b 	   E       	   F       	   G      
, 	   H      "C 	   I      RZ 	   J      t 	   K      ڋ 	   L      2 	   M      2 	   N       	   O       	   P      " 	   Q      z% 	   R      < 	   S      W 	   T      2n 	   U       	   V      ڠ 	   W      R 	   X       	   Y       	   Z       	   [       	   \      B0 	   ]      rG 	   ^      ^ 	   _      u 	   `       	   a      ң 	   b      Һ 	   c       	   d       	   e      2  	   f       	   g      r0 	   h      I 	   i      ` 	   j      B} 	   k      B 	   l       	   m       	   n      B 	   o       	   p       	   q      B. 	   r      rE 	   s      a 	   t      y 	   u      2 	   v      b 	   w       	   x       	   y       	   z      " 	   {      R 	   |      2 	   }      I 	   ~      ` 	         x 	         B 	          	         ҽ 	          	         b 	          	          	         
2 	         "I 	         R` 	         y 	          	         b 	         b 	          	          	         R 	         ) 	         @ 	         2[ 	         br 	         ڍ 	         
 	          	          	          	           	         B  	         r4  	         K  	         b  	         z  	         2  	         b  	           	           	           	         "! 	         ! 	         5! 	         2O! 	         2f! 	         ! 	         ! 	         ! 	         b! 	         ! 	          " 	         2" 	         2" 	         I" 	         Re" 	         |" 	         " 	         " 	         " 	         B" 	         r" 	         # 	         # 	         6# 	         2M# 	         bd# 	         {# 	         # 	         
# 	         R# 	         # 	         # 	         $ 	         $ 	         *5$ 	         BL$ 	         rc$ 	         |$ 	         $ 	         $ 	         $ 	         $ 	         $ 	         B% 	         .% 	         E% 	         "`% 	         Rw% 	         ʒ% 	         % 	         r% 	         % 	         % 	         & 	         2"& 	         b9& 	         P& 	         g& 	         ~& 	         "& 	         R& 	         & 	         & 	         & 	         
' 	         B!' 	         r8' 	         O' 	         f' 	         ' 	         ' 	         b' 	         ' 	         ' 	         B( 	         r( 	         4( 	         L( 	         g( 	         ~( 	         ( 	         "( 	         R( 	         ( 	         ( 	         	) 	         !) 	         B8) 	         rO) 	         f) 	         }) 	          ) 	         J) 	         ) 	         ) 	         ") 	         j	* 	          * 	         7* 	         BO* 	   	      rf* 	   
      * 	         * 	         * 	         * 	         * 	         * 	         B+ 	         0+ 	         G+ 	         Ra+ 	         x+ 	         + 	         *+ 	         + 	         + 	         + 	         2, 	         #, 	         2:, 	         bQ, 	         h, 	         , 	          , 	   !      ", 	   "      R, 	   #      , 	   $      , 	   %      
- 	   &      "- 	   '      B9- 	   (      rP- 	   )      rg- 	   *      Z- 	   +      Z- 	   ,      - 	   -      J- 	   .      z- 	   /      - 	   0      . 	   1      z,. 	   2      C. 	   3      "_. 	   4      Rv. 	   5      . 	   6      . 	   7      . 	   8      . 	   9      B. 	   :      r/ 	   ;      / 	   <      // 	   =      G/ 	   >      2^/ 	   ?      bu/ 	   @      / 	   A      ڣ/ 	   B      "/ 	   C      j/ 	   D      / 	   E       0 	   F      B0 	   G      /0 	   H      F0 	   I      ^0 	   J      2u0 	   K      b0 	   L      0 	   M      z0 	   N      0 	   O      0 	   P      "1 	   Q      %1 	   R      <1 	   S      2V1 	   T      bm1 	   U      ڈ1 	   V      
1 	   W      1 	   X      1 	   Y      1 	   Z      2 	   [      B2 	   \      r/2 	   ]      F2 	   ^      ]2 	   _      u2 	   `      22 	   a      b2 	   b      2 	   c      2 	   d      2 	   e      " 3 	   f      R3 	   g      .3 	   h      E3 	   i      \3 	   j      t3 	   k      3 	   l      3 	   m      ڽ3 	   n      J3 	   o      3 	   p      4 	   q      &4 	   r      =4 	   s      2Y4 	   t      bp4 	   u      4 	   v      4 	   w      4 	   x      "4 	   y      R4 	   z      4 	   {      5 	   |      )5 	   }      A5 	   ~      BX5 	         ro5 	         5 	         5 	         25 	         z5 	         5 	         
5 	         R6 	         )6 	         @6 	         X6 	         Bo6 	         r6 	         6 	         6 	         r6 	         Z6 	         6 	         b7 	         -7 	         G7 	         2^7 	         bu7 	         7 	         £7 	         7 	         "7 	         R7 	          8 	         8 	         .8 	         F8 	         B]8 	         rt8 	         8 	         Ң8 	         8 	         28 	         b8 	         8 	         9 	         -9 	         D9 	         \9 	         s9 	         9 	         29 	         9 	         9 	         9 	         : 	         r(: 	         C: 	         b_: 	         v: 	         : 	         : 	         »: 	         : 	         ": 	         R; 	         ; 	         /; 	         F; 	         ^; 	        Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })                                b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                                 #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        ;         <  Y         Z  w         x                                    !  6          X  m                                              8  T          q                                    C          D   j!         k!  "         "  #                                    b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                            b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                                 3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            %         '  L         N  s         u                                              6	         8	  ]
         _
                                                        "  G         I  n         p                                   
           1         3  X         Z                                                         B          D   i!         k!  "          $  2%          L%  f&          &  '          '  (          (  *         K  L         L  N         N  <O         >O  jP         lP  Q         Q  R         R  S         S  "U         $U  PV         RV  ~W         W  X         X  Y         Y  [         [  .\         0\  Z]         \]  ^         ^  _         _  `         `  
b         b  6c         8c  `d         bd  e         e  f         f  g         g  i         
i  2j         4j  \k         ^k  l         l  m         m  n         ȑ                        3         4  W         X  {         |             Ù         ę                        /         0  S         T  w         x                                              +         ,  O         P  s         t                        ߪ                      '         (  K         L  o         p                        ۲         ܲ                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               a!         b!  2         2  =                                    b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1                    	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     O
            
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                 
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            <     -,-p     ,-U-     U--    -4-    4-uC-    uC-v-t    v-30-     30-7-X    7--    -ub-p	    ub-߂-x
    ߂--
    -U-    U-ճ-\    ճ-׳-<                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     u            !          R!  1          1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        7  M          n                                                6  9          Y  \	          |	  
          
                                              5         6  R                      '         (  :         ;  M         N  `                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        T$  w%         x%  &         vL  M         M  N         N  O         O  Q         Q  (R         *R  LS         NS  pT         rT  U         U  V         V  W         W   Y         Y  $Z         &Z  H[         J[  l\         n\  ]         ]  ^         ^  _         _  `         `   b         "b  Dc         Fc  hd         jd  e         e  f         f  g         g  h         h  j         j  @k         Bk  dl         fl  m         m  n         n  o         o  p         p  r         r  <s         >s  `t         bt  u         u  v         v  w         w  x         x  z         z  8{         :{  \|         ^|  }         }  ~         ~                                   4         6  X         Z  |         ~             ć         Ƈ                        0         2  T         V  x         z                                   	         
  -         .  Q          R  W          v  {                        Ø                       )         *  M         N  q         r                        ݠ         ޠ             %         &  I         J  m         n                        ٨         ڨ             !         "  E         F  i         j                        հ                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               O!         P!  1         1  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    Z
         [
             "                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    Z
         [
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-a-     a--    -1<-x    1<-J-|    J-~-\    ~-]3-    ]3--h    -b-    b-	n-	    	n--h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     |            (!          X!  1          2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        9         :  V         W  s         t                                      2          S  i                                              3  P          l                                    @          A   h!         i!  "         "  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            &         '  M         N  t         u                                              7	         8	  ^
         _
                                              !         "  H         I  o         p                                              2         3  Y         Z                                                         C          D   j!         k!  "          *  6+          P+  j,          ,  -          -  .          .  0         K  L         L  N         N  =O         >O  kP         lP  Q         Q  R         R  S         S  #U         $U  QV         RV  W         W  X         X  Y         Y  [         [  /\         0\  []         \]  ^         ^  _         _  `         `  b         b  7c         8c  ad         bd  e         e  f         f  g         g  	i         
i  3j         4j  ]k         ^k  l         l  m         m  n         ȑ                        2         4  V         X  z         |                      ę             
           .         0  R         T  v         x                                              *         ,  N         P  r         t                        ު                      &         (  J         L  n         p                        ڲ         ܲ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               e!         f!  2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     N
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,-U-     U--    -<-x    <-J-|    J-}-\    }-]3-    ]3--h    -b-    b-	n-	    	n--h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     t            !          P!  1          1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        9         :  V         W  s         t                                      (          O  a                                              /  D          f  {                                  ,          -   J!         K!  h"         i"  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                      '  G         N  n         u                                   
           1	         8	  X
         _
                                                       "  B         I  i         p                                              ,         3  S         Z  z                                                       =          D   d!         k!  "           0  51          T1  i2          2  3          3  4          4  6         K  L         L  N         N  5O         >O  cP         lP  Q         Q  R         R  S         S  U         $U  IV         RV  wW         W  X         X  Y         Y  Z         [  '\         0\  S]         \]  ^         ^  _         _  `         `  b         b  /c         8c  Yd         bd  e         e  f         f  g         g  i         
i  +j         4j  Uk         ^k  l         l  m         m  n         ȑ                        0         4  T         X  x         |                      ę                        ,         0  P         T  t         x                                              (         ,  L         P  p         t                        ܪ                       $         (  H         L  l         p                        ز         ܲ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               R!         V!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     H
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,-s-     s--    -5-x    5-C-|    C-v-\    v--    --h    -Kb-    Kb-n-	    n--h
    -׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               >!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        9  U          r                                       $          >  E          _  f	          	  
          
                                   6         7  ]         ^           *  =         >  Q         R  e         f  y         z                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            #         %  H         J  m         o                                              &	         (	  K
         M
  p         r                                              )         +  N         P  s         u                                            	  ,         .  Q         S  v         x                                   
             /!         1!  T"          /  1          51  [2          x2  3          3  4          4  $6         I  J         J  K         K  M         M  &N         (N  KO         MO  pP         rP  Q         Q  R         R  S         S  U         U  )V         +V  NW         PW  sX         uX  Y         Y  Z         Z  [         [  ]         	]  ,^         .^  Q_         S_  v`         x`  a         a  b         b  c         c  e         e  4f         6f  Zg         \g  h         h  i         i  j         j  k         k  m         m  Bn         Dn  jo         lo  p         p  q         q  r         r  
t         t  2u         4u  Zv         \v  w         w  x         x  y         y  z         z  "|         $|  J}         L}  r~         t~                      Ā                        :         <  b         d                        ڈ         ܈             *         ,  R         T  z         |                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     {            #!          P!  1          1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     O
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-U-     U--    -<-x    <-J-|    J-}-\    }-]3-    ]3--h    -b-    b-	n-	    	n--h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     u            !          P!  1          1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        ;         <  Y         Z  w         x                                      .          T  g                                              4  J          k                                    /          0   L!         M!  i"         j"  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            !         '  H         N  o         u                                              2	         8	  Y
         _
                                                       "  C         I  j         p                                              -         3  T         Z  {                                                       >          D   e!         k!  "          $6  97          X7  m8          8  9          9  :          :  	<         K  L         L  N         N  4O         >O  bP         lP  Q         Q  R         R  S         S  U         $U  HV         RV  vW         W  X         X  Y         Y  Z         [  &\         0\  R]         \]  ~^         ^  _         _  `         `  b         b  .c         8c  Xd         bd  e         e  f         f  g         g   i         
i  *j         4j  Tk         ^k  ~l         l  m         m  n         ȑ                        1         4  U         X  y         |                      ę             	           -         0  Q         T  u         x                                              )         ,  M         P  q         t                        ݪ                      %         (  I         L  m         p                        ٲ         ܲ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               T!         Z!  2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     G
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,-s-     s--    -5-x    5-C-|    C-v-\    v--    --l    -Kb-    Kb-n-     n--     -׳-                    =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               >!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        9  V          r                                      #          >  D          _  e	          	  
          
                                   9         :  a         b           /  C         D  X         Y  m         n                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7         operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1                    >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     |            $!          P!  1          1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1                   |            $         %  I         J  n         o                                              '	         (	  L
         M
  q         r                                              *         +  O         P  t         u                                            	  -         .  R         S  w         x                                                0!         1!  U"          A6  g7          7  8          8  9          
:  0;          M;  s<         I  J         J  K         K  M         M  'N         (N  LO         MO  qP         rP  Q         Q  R         R  S         S  U         U  *V         +V  OW         PW  tX         uX  Y         Y  Z         Z  [         [  ]         	]  -^         .^  R_         S_  w`         x`  a         a  b         b  c         c  e         e  5f         6f  [g         \g  h         h  i         i  j         j  k         k  m         m  Cn         Dn  ko         lo  p         p  q         q  r         r  t         t  3u         4u  [v         \v  w         w  x         x  y         y  z         z  #|         $|  K}         L}  s~         t~             À         Ā                        ;         <  c         d                        ۈ         ܈             +         ,  S         T  {         |              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     N
            
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                           b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-m-     m--    -4-x    4-C-|    C-v-\    v--    --h    -Eb-    Eb-	n-	    	n--h
    -!׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               7!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        7  L          n                                                6  :          Y  ]	          |	  
          
                                              9         :  W                      ,         -  ?         @  R         S  e                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                     %  C         J  h         o                                              !	         (	  F
         M
  k         r                                              $         +  I         P  n         u                                            	  '         .  L         S  q         x                                                *!         1!  O"          <  =          =  >          ?  7@          Y@  zA          A  B         I  J         J  K         K  L         M  !N         (N  FO         MO  kP         rP  Q         Q  R         R  S         S  T         U  $V         +V  IW         PW  nX         uX  Y         Y  Z         Z  [         [  ]         	]  '^         .^  L_         S_  q`         x`  a         a  b         b  c         c  	e         e  /f         6f  Ug         \g  {h         h  i         i  j         j  k         k  m         m  ;n         Dn  co         lo  p         p  q         q  r         r  t         t  +u         4u  Sv         \v  {w         w  x         x  y         y  z         z  |         $|  C}         L}  k~         t~                      Ā                        3         <  [         d                        ӈ         ܈             #         ,  K         T  s         |                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     t            !          L!  1          1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     H
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     --    -:-x    :-oI-|    oI-|-\    |-y-    y--h    -Wb-    Wb-x-	    x--h
    -!׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               B!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P           %         &  8         9  K	         L	  ^
         _
  q         !  >         ?  \         ]  z         {                                                                        C  =          f  {                                      !          B!  W"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                        ?         @  _         `                                                        	  
          
  ?         @  _         `                                                                      ?         @  _         `                                                                      ?         @  _         `                          !         !  "         "  $         $  )%         +%  L&         N&  o'         q'  (         (  )         )  *         *  +         +  -          -  A.         C.  d/         f/  0         0  1         1  2         2  3         3  5         5  66         86  Y7         [7  |8         ~8  9         9  :         :  ;         ;  =         
=  +>         ->  N?         P?  q@         s@  A         A  B         B  C         "D  CE         EE  fF         hF  G         G  H         I  K         K  $L         &L  GM         IM  jN         lN  O         O  P         P  Q         Q  R         R  T         T  <U         >U  _V         aV  W         W  X         X  Y         Y  Z         Z  \         \  1]         3]  T^         V^  w_         y_  `         `  a         a  b         b  d         d  &e         (e  If         Kf  lg          w  u                      ǅ  ņ                                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     {            &!          V!  1          2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-m-     m--    -4-x    4-{C-|    {C-v-\    v--    --h    -Eb-    Eb-	n-	    	n--h
    -!׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               9!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        7  M          n                                                6  9          Y  \	          |	  
          
                                              5         6  R                      '         (  :         ;  M         N  `                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                     %  D         J  i         o                                              "	         (	  G
         M
  l         r                                               %         +  J         P  o         u                                            	  (         .  M         S  r         x                                                +!         1!  P"          B   D          "D  CE          eE  F          F  G          G  I         I  J         J  K         K  L         M  "N         (N  GO         MO  lP         rP  Q         Q  R         R  S         S   U         U  %V         +V  JW         PW  oX         uX  Y         Y  Z         Z  [         [  ]         	]  (^         .^  M_         S_  r`         x`  a         a  b         b  c         c  
e         e  0f         6f  Vg         \g  |h         h  i         i  j         j  k         k  m         m  :n         Dn  bo         lo  p         p  q         q  r         r  t         t  *u         4u  Rv         \v  zw         w  x         x  y         y  z         z  |         $|  B}         L}  j~         t~                      Ā             
           2         <  Z         d                        ҈         ܈             "         ,  J         T  r         |                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     u            !          L!  1          1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     G
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-y-     y--    -:-x    :-oI-|    oI-|-\    |-C-    C--h    -cb-    cb-x-	    x--h
    -!׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               >!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P           %         &  8         9  K	         L	  ^
         _
  q         !  =         >  Z         [  w         x                                                                       >  7          a  w                                     !          =!  S"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                        >         @  ^         `  ~                                                      	  
          
  >         @  ^         `  ~                                                                    >         @  ^         `  ~                                                                    >         @  ^         `  ~                        !         !  "         "  $         $  *%         +%  M&         N&  p'         q'  (         (  )         )  *         *  +         +  -          -  B.         C.  e/         f/  0         0  1         1  2         2  3         3  5         5  76         86  Z7         [7  }8         ~8  9         9  :         :  ;         ;  	=         
=  ,>         ->  O?         P?  r@         s@  A         A  B         B  C         "D  DE         EE  gF         hF  G         G  H         I  K         K  %L         &L  HM         IM  kN         lN  O         O  P         P  Q         Q  R         R  T         T  =U         >U  `V         aV  W         W  X         X  Y         Y  Z         Z  \         \  2]         3]  U^         V^  x_         y_  `         `  a         a  b         b  d         d  'e         (e  Jf         Kf  mg          ?  =          g  e                                  ߍ  ݎ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     |            )!          Z!  1          2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-y-     y--    -:-x    :-iI-|    iI-|-\    |--    --h    -Kb-    Kb-x-	    x--h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ;!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P           /         0  B         C  U	         V	  h
         i
  {         !  =         >  Z         [  w         x                                                                       6  )          W  i                                     !          ;!  M"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                        <         @  \         `  |                                                      	  
          
  <         @  \         `  |                                                                    <         @  \         `  |                                                                    <         @  \         `  |                        !         !  "         "  $         $  $%         +%  G&         N&  j'         q'  (         (  )         )  *         *  +         +  -          -  <.         C.  _/         f/  0         0  1         1  2         2  3         3  5         5  16         86  T7         [7  w8         ~8  9         9  :         :  ;         ;  =         
=  &>         ->  I?         P?  l@         s@  A         A  B         B  C         "D  >E         EE  aF         hF  G         G  H         I  J         K  L         &L  BM         IM  eN         lN  O         O  P         P  Q         Q  R         R  T         T  7U         >U  ZV         aV  }W         W  X         X  Y         Y  Z         Z  	\         \  ,]         3]  O^         V^  r_         y_  `         `  a         a  b         b  c         d  !e         (e  Df         Kf  gg          k  l          m  n          Dn  =o          lo  ep          p  q                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     t            !          N!  1          1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 }
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-5-x    5-C-|    C-v-\    v-
-    
--h    -ib-    ib-x-	    x--h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               D!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        9  U          r                                       $          >  E          _  f	          	  
          
                                   6         7  ]         ^           *  =         >  Q         R  e         f  y         z                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                    8  7          T  S          p  o         e$  %         %  &         &  '         '  (         (  *         
*  )+         ++  J,         L,  k-         m-  .         .  /         /  0         0  1         1  3         3  14         34  R5         T5  s6         u6  7         7  8         8  9         9  :         :  <         <  9=         ;=  Z>         \>  {?         }?  @         @  A         A  B         B  C         D   E         "E  AF         F  G         G  I         
I  )J         +J  JK         LK  kL         mL  M         M  N         N  O         O  P         P  R         R  1S         3S  RT         TT  sU         uU  V         V  W         W  X         X  Y         Y  [         [  9\         ;\  Z]         \]  {^         }^  _         _  `         `  a         a  b         c   d         "d  Ae         Ce  bf         df  g         g  h         h   j         "j  Dk         Fk  hl         jl  m         m  n         n  o         o  p         p  r         r  @s         Bs  dt         ft  u         u  v         v  w         w  x         x  z         z  <{         >{  `|         b|  }         }  ~         ~                                   8         :  \         ^                        ȇ         ʇ                        4                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               <!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-s-     s--    -:-x    :-uI-|    uI-|-\    |--    --h    -Qb-    Qb-x-	    x--h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               9!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P           /         0  B         C  U	         V	  h
         i
  {         &  C         D  a         b                                                                                  @  2          a  t                                     !          E!  X"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                        =         @  ]         `  }                                                      	  
          
  =         @  ]         `  }                                                                    =         @  ]         `  }                                                                    =         @  ]         `  }                        !         !  "         "  $         $  %%         +%  H&         N&  k'         q'  (         (  )         )  *         *  +         +  -          -  =.         C.  `/         f/  0         0  1         1  2         2  3         3  5         5  26         86  U7         [7  x8         ~8  9         9  :         :  ;         ;  =         
=  '>         ->  J?         P?  m@         s@  A         A  B         B  C         "D  ?E         EE  bF         hF  G         G  H         I  J         K   L         &L  CM         IM  fN         lN  O         O  P         P  Q         Q  R         R  T         T  8U         >U  [V         aV  ~W         W  X         X  Y         Y  Z         Z  
\         \  -]         3]  P^         V^  s_         y_  `         `  a         a  b         b  c         d  "e         (e  Ef         Kf  hg          q  r          r  s          t  u          4u  -v          \v  Uw                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     u            !          R!  1          1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 }
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     --    -5-x    5-C-|    C-w-\    w-
-    
--h    -ib-    ib-x-	    x--h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               @!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        9  V          r                                      #          >  D          _  e	          	  
          
                                   9         :  a         b           /  C         D  X         Y  m         n                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                               	          	  
         e$  %         %  &         &  '         '  (         (  	*         
*  *+         ++  K,         L,  l-         m-  .         .  /         /  0         0  1         1  3         3  24         34  S5         T5  t6         u6  7         7  8         8  9         9  :         :  <         <  :=         ;=  [>         \>  |?         }?  @         @  A         A  B         B   D         D  !E         "E  BF         F  G         G  	I         
I  *J         +J  KK         LK  lL         mL  M         M  N         N  O         O  P         P  R         R  2S         3S  ST         TT  tU         uU  V         V  W         W  X         X  Y         Y  [         [  :\         ;\  []         \]  |^         }^  _         _  `         `  a         a   c         c  !d         "d  Be         Ce  cf         df  g         g  h         h  !j         "j  Ek         Fk  il         jl  m         m  n         n  o         o  p         p  r         r  As         Bs  et         ft  u         u  v         v  w         w  x         x  z         z  ={         >{  a|         b|  }         }  ~         ~                                   9         :  ]         ^                        ɇ         ʇ                        5                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               <!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     --    -4-x    4-C-|    C-v-\    v-
-    
--h    -cb-    cb-x-	    x--h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               =!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        7  L          n                                                6  :          Y  ]	          |	  
          
                                              9         :  W                      ,         -  ?         @  R         S  e                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                     4  .          P  J          l  f                     e$  %         %  &         &  '         '  (         (  *         
*  $+         ++  E,         L,  f-         m-  .         .  /         /  0         0  1         1  3         3  ,4         34  M5         T5  n6         u6  7         7  8         8  9         9  :         :  <         <  4=         ;=  U>         \>  v?         }?  @         @  A         A  B         B  C         D  E         "E  <F         F  G         G  I         
I  $J         +J  EK         LK  fL         mL  M         M  N         N  O         O  P         P  R         R  ,S         3S  MT         TT  nU         uU  V         V  W         W  X         X  Y         Y  [         [  4\         ;\  U]         \]  v^         }^  _         _  `         `  a         a  b         c  d         "d  <e         Ce  ]f         df  ~g         g  h         h  j         "j  =k         Fk  al         jl  m         m  n         n  o         o  p         p  r         r  9s         Bs  ]t         ft  u         u  v         v  w         w  x         x  z         z  5{         >{  Y|         b|  }}         }  ~         ~                                   1         :  U         ^  y                               ʇ             	           -                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               5!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 }
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-m-     m--    -:-x    :-{I-|    {I-|-\    |-+-    +--h    -ob-    ob-'-	    '--h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               8!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P           %         &  8         9  K	         L	  ^
         _
  q         !  >         ?  \         ]  z         {                                                                        C  =          f  {                                      !          B!  W"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       7         8  S         T  o         p                                              	         	             3         4  O         P  k         l                                                                    /         0  K         L  g         h                                                                    +          ,   G!         !  "         "  #         #  %         %  :&         <&  Y'         ['  x(         z(  )         )  *         *  +         +  ,         ,  .         .  2/         4/  Q0         S0  p1         r1  2         2  3         3  4         4  5         5  7         7  *8         ,8  I9         K9  h:         j:  ;         ;  <         <  =         =  >         >  @         @  "A         $A  AB         CB  `C         C  E         E  4F         6F  SG         UG  rH         tH  I          V  W          W  Y          $Y  BZ          dZ  [          [  \         (  E         G  d         f                                 Õ                                    =         ?  \         ^  {         }                        ؟         ڟ                        5         7  T         V  s         u                        Ш         Ҩ                        -                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ;!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   [      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-y-     y--    -4-x    4-C-|    C-v-\    v-
-    
--h    -cb-    cb-x-	    x--h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ;!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        7  M          n                                                6  9          Y  \	          |	  
          
                                              5         6  R                      '         (  :         ;  M         N  `                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                                                    e$  %         %  &         &  '         '  (         (  *         
*  %+         ++  F,         L,  g-         m-  .         .  /         /  0         0  1         1  3         3  -4         34  N5         T5  o6         u6  7         7  8         8  9         9  :         :  <         <  5=         ;=  V>         \>  w?         }?  @         @  A         A  B         B  C         D  E         "E  =F         F  G         G  I         
I  %J         +J  FK         LK  gL         mL  M         M  N         N  O         O  P         P  R         R  -S         3S  NT         TT  oU         uU  V         V  W         W  X         X  Y         Y  [         [  5\         ;\  V]         \]  w^         }^  _         _  `         `  a         a  b         c  d         "d  =e         Ce  ^f         df  g         g  h         h  j         "j  <k         Fk  `l         jl  m         m  n         n  o         o  p         p  r         r  8s         Bs  \t         ft  u         u  v         v  w         w  x         x  z         z  4{         >{  X|         b|  |}         }  ~         ~                                   0         :  T         ^  x                               ʇ                        ,                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               7!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 }
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-s-     s--    -:-x    :-uI-|    uI-|-\    |-+-    +--h    -ib-    ib-'-	    '--h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               :!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P           %         &  8         9  K	         L	  ^
         _
  q         !  =         >  Z         [  w         x                                                                       >  7          a  w                                     !          =!  S"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       6         8  R         T  n         p                                              	         	             2         4  N         P  j         l                                                                    .         0  J         L  f         h                                                                    *          ,   F!         !  "         "  #         #  %         %  ;&         <&  Z'         ['  y(         z(  )         )  *         *  +         +  ,         ,  .         .  3/         4/  R0         S0  q1         r1  2         2  3         3  4         4  5         5  7         7  +8         ,8  J9         K9  i:         j:  ;         ;  <         <  =         =  >         >  @         @  #A         $A  BB         CB  aC         C  E         E  5F         6F  TG         UG  sH         tH  I          \  ^          $^  B_          d_  `          `  a          a  c         (  F         G  e         f                                 Õ                                     >         ?  ]         ^  |         }                        ٟ         ڟ                        6         7  U         V  t         u                        Ѩ         Ҩ                        .                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               7!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   \      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-g-     g--    -:-x    :-uI-|    uI-|-\    |-+-    +--h    -ib-    ib---	    ---h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               1!         @!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P           /         0  B         C  U	         V	  h
         i
  {         !  =         >  Z         [  w         x                                                                       6  )          W  i                                     !          ;!  M"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       4         8  P         T  l         p                                              	         	             0         4  L         P  h         l                                                                    ,         0  H         L  d         h                                                                    (          ,   D!         !  "         "  #         #  %         %  5&         <&  T'         ['  s(         z(  )         )  *         *  +         +  ,         ,  .         .  -/         4/  L0         S0  k1         r1  2         2  3         3  4         4  5         5  7         7  %8         ,8  D9         K9  c:         j:  ;         ;  <         <  =         =  >         >  ?         @  A         $A  <B         CB  [C         C  E         E  /F         6F  NG         UG  mH         tH  I          $c  =d          dd  }e          e  f          f  g          $h  =i         (  @         G  _         f  ~                               Õ  ۖ                                  8         ?  W         ^  v         }                        ӟ         ڟ                        0         7  O         V  n         u                        ˨         Ҩ             	           (                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               4!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   \      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--      -/-*     /-C<-Z     C<-J-z     J-+~-R     +~--     --     -]b-     ]b---     ---     --׳-                    =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               H!         R!  1         1  =         z   input0Elements                                b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                            b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                                 #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        ;         <  Y         Z  w         x                                    !  6          X  m                                              8  T          q                                    C          D   j!         k!  "         "  #                                    b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                            b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                                 3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       :         <  X         Z  v         x                                              
         
  *         ,  H         J  f         h                                                                    8         :  V         X  t         v                                              
           (         *  F         H  d          f   !         #  $         $  &         &  :'         <'  Z(         \(  z)         |)  *         *  +         +  ,         ,  -         -  /         /  :0         <0  Z1         \1  z2         |2  3         3  4         4  5         5  6         6  8         8  :9         <9  Z:         \:  z;         |;  <         <  =         =  >         >  ?         ?  A         A  :B         <B  ZC         \C  zD         |D  E         F  G         G  H         H  J         J  ?K         @K  eL         fL  M         M  N         N  O         O  P         P  #R         $R  ES         GS  hT         jT  U         U  V         V  W         W  X         X  Z         Z  :[         <[  ]\         _\  ]         ]  ^         ^  _         _  `         `  b         b  %c         'c  Fd         Hd  ge         ie  f         f  g         g  h                      5  V          x              ܇                                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               4!         >!  1         1  =                                    b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1                    	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   [            
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                           b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-m-     m--    -:-x    :-I-|    I-|-\    |-+-    +--h    -cb-    cb---	    ---h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               5!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P           /         0  B         C  U	         V	  h
         i
  {         &  C         D  a         b                                                                                  @  2          a  t                                     !          E!  X"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       5         8  Q         T  m         p                                              	         	             1         4  M         P  i         l                                                                    -         0  I         L  e         h                                                                    )          ,   E!         !  "         "  #         #  %         %  6&         <&  U'         ['  t(         z(  )         )  *         *  +         +  ,         ,  .         .  ./         4/  M0         S0  l1         r1  2         2  3         3  4         4  5         5  7         7  &8         ,8  E9         K9  d:         j:  ;         ;  <         <  =         =  >         >  ?         @  A         $A  =B         CB  \C         C  E         E  0F         6F  OG         UG  nH         tH  I          di  }j          j  k          k  l          $m  =n          dn  }o         (  A         G  `         f                                 Õ  ܖ                                  9         ?  X         ^  w         }                        ԟ         ڟ                        1         7  P         V  o         u                        ̨         Ҩ             
           )                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               2!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   ]      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-C<-x    C<-J-|    J-+~-\    +~--    --h    -Wb-    Wb---	    ---h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               F!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        9         :  V         W  s         t                                      2          S  i                                              3  P          l                                    @          A   h!         i!  "         "  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       ;         <  Y         Z  w         x                                              
         
  +         ,  I         J  g         h                                                                    9         :  W         X  u         v                                                         )         *  G         H  e          f   !         #  $         $  &         &  ;'         <'  [(         \(  {)         |)  *         *  +         +  ,         ,  -         -  /         /  ;0         <0  [1         \1  {2         |2  3         3  4         4  5         5  6         6  8         8  ;9         <9  [:         \:  {;         |;  <         <  =         =  >         >  ?         ?  A         A  ;B         <B  [C         \C  {D         |D  E         F  G         G  H         H  J         J  >K         @K  dL         fL  M         M  N         N  O         O  P         P  "R         $R  FS         GS  iT         jT  U         U  V         V  W         W  X         X  Z         Z  ;[         <[  ^\         _\  ]         ]  ^         ^  _         _  `         `  b         b  &c         'c  Gd         Hd  he         ie  f         f  g         g  h          A  b                      ǋ            
  +          M  n                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               2!         >!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   \      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-<-x    <-J-|    J-~-\    ~--    --h    -Wb-    Wb---	    ---h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               I!         V!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        9         :  V         W  s         t                                      (          O  a                                              /  D          f  {                                  ,          -   J!         K!  h"         i"  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       5         <  S         Z  q         x                                              
         
  %         ,  C         J  a         h                                                                    3         :  Q         X  o         v                                                         #         *  A         H  _          f   }!         #  $         $  &         &  3'         <'  S(         \(  s)         |)  *         *  +         +  ,         ,  -         -  /         /  30         <0  S1         \1  s2         |2  3         3  4         4  5         5  6         6  8         8  39         <9  S:         \:  s;         |;  <         <  =         =  >         >  ?         ?  A         A  3B         <B  SC         \C  sD         |D  E         F  G         G  H         H  J         J  <K         @K  bL         fL  M         M  N         N  O         O  P         P   R         $R  @S         GS  cT         jT  U         U  V         V  W         W  X         X  Z         Z  5[         <[  X\         _\  {]         ]  ^         ^  _         _  `         `  a         b   c         'c  Ad         Hd  be         ie  f         f  g         g  h          h  i          i  k          ,k  8l          \l  hm          m  n                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               -!         :!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   \      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -5-    5-:-    :-I-    I-|-d    |--    --0    --    -b-H	    b-c-P
    c-s-
    s-3׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               I!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P           %         &  8         9  K	         L	  ^
         _
  q         !  >         ?  \         ]  z         {                                                                        C  =          f  {                                      !          B!  W"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q   w  operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       \                        ?         @  _         `                                                        	  
          
  ?         @  _         `                                                                      ?         @  _         `                                                                      ?         @  _         `                           !          !  "          "  #           $  %           %  !&         @&  ^'         `'  ~(         (  )         )  *         *  +         +  ,          -  .          .  >/         @/  ^0         `0  ~1         1  2         2  3         3  4         4  5          6  7          7  >8         @8  ^9         `9  ~:         :  ;         ;  <         <  =         =  >          ?  @          @  >A         G  
I         I  *J         +J  JK         KK  jL         kL  M         M  N         N  O         O  P         P  
R         R  *S         +S  JT         KT  jU         kU  V         V  W         W  X         X  Y         Y  
[         [  *\         +\  J]         K]  j^         k^  _         _  `         `  a         a  b         b  d         d  6e         7e  Xf         Yf  zg         {g  h                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               I!         \!  1         
2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-<-x    <-J-|    J-~-\    ~--    --h    -Wb-    Wb-3-	    3--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               I!         V!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        ;         <  Y         Z  w         x                                      .          T  g                                              4  J          k                                    /          0   L!         M!  i"         j"  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       6         <  T         Z  r         x                                              
         
  &         ,  D         J  b         h                                                                    4         :  R         X  p         v                                                         $         *  B         H  `          f   ~!         #  $         $  &         &  2'         <'  R(         \(  r)         |)  *         *  +         +  ,         ,  -         -  /         /  20         <0  R1         \1  r2         |2  3         3  4         4  5         5  6         6  8         8  29         <9  R:         \:  r;         |;  <         <  =         =  >         >  ?         ?  A         A  2B         <B  RC         \C  rD         |D  E         F  G         G  H         H  J         J  =K         @K  cL         fL  M         M  N         N  O         O  P         P  !R         $R  AS         GS  dT         jT  U         U  V         V  W         W  X         X  Z         Z  6[         <[  Y\         _\  |]         ]  ^         ^  _         _  `         `   b         b  !c         'c  Bd         Hd  ce         ie  f         f  g         g  h          n  o          o  p          q  (r          Lr  Xs          |s  t                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               -!         :!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   ]      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -/-    /-:-    :-{I-    {I-|-d    |--    --0    --    -b-H	    b-c-P
    c-s-
    s-3׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               F!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P           %         &  8         9  K	         L	  ^
         _
  q         !  =         >  Z         [  w         x                                                                       >  7          a  w                                     !          =!  S"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q   w  operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       \                        >         @  ^         `  ~                                                      	  
          
  >         @  ^         `  ~                                                                    >         @  ^         `  ~                                                                    >         @  ^         `  ~                        !         !  "         "  #          $  %          %  ?&          @&  A'          `'  a(          (  )          )  *          *  +         +  ,          -  .          .  ?/         @/  _0         `0  1         1  2         2  3         3  4         4  5          6  7          7  ?8         @8  _9         `9  :         :  ;         ;  <         <  =         =  >          ?  @          @  ?A         G  	I         I  )J         +J  IK         KK  iL         kL  M         M  N         N  O         O  P         P  	R         R  )S         +S  IT         KT  iU         kU  V         V  W         W  X         X  Y         Y  	[         [  )\         +\  I]         K]  i^         k^  _         _  `         `  a         a  b         b  d         d  5e         7e  Wf         Yf  yg         {g  h                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               E!         X!  1         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -A-    A-:-    :-{I-    {I-|-d    |--    --0    --    -b-H	    b-c-P
    c-y-
    y-3׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               J!         X!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P           /         0  B         C  U	         V	  h
         i
  {         !  =         >  Z         [  w         x                                                                       6  )          W  i                                     !          ;!  M"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q   w  operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       \                        <         @  \         `  |                                                      	  
          
  <         @  \         `  |                                                                    <         @  \         `  |                                                                    <         @  \         `  |                        !         !  "         "  #          $  %          %  9&         @&  Y'         `'  y(         (  )         )  *         *  +          +  ,           -  -           .  /          @/  <0          `0  \1         1  2         2  3         3  4         4  5          6  7          7  98         @8  Y9         `9  y:         :  ;         ;  <         <  =         =  >          ?  @          @  9A         G  I         I  'J         +J  GK         KK  gL         kL  M         M  N         N  O         O  P         P  R         R  'S         +S  GT         KT  gU         kU  V         V  W         W  X         X  Y         Y  [         [  '\         +\  G]         K]  g^         k^  _         _  `         `  a         a  b         b  d         d  3e         7e  Uf         Yf  wg         {g  h                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               B!         X!  1         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -/-    /-y<-    y<-K-    K-a~-d    a~--    --p    -b-    b--
    -)-
    )--t    -3׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               F!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        ;         <  Y         Z  w         x                                    !  6          X  m                                              8  O          q                         %         &  U          V   !         !  "         "  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         |                                                           $  8%         :%  [&         \&  }'         ~'  (         (  )         )  *         *  ,         ,  '-         (-  I.         J.  k/         l/  0         0  1         1  2         2  3         3  5         5  76         86  Y7         Z7  {8         |8  9         9  :         :  ;         ;  =         =  %>         &>  G?         H?  i@         j@  A         A  B         B  C         C  D         D  F         F  5G         6G  VH         XH  xI         zI  J         J  K         K  L         L  "N         $N  UO         WO  P         P  Q         Q  R         R  !T         #T  TU         VU  V         V  W         W  X         X  'Z         (Z  ][         ^[  \         \  ]         ]  ^          _  5`         6`  ka         la  b         b  c         c  e         e  ?f         @f  sg         tg  h         h  i         i  k         k  Cl         Dl  wm         xm  n         n  o         o   q         q  "r         $r  Ds         Fs  ft         ht  u         u  v         v  w         w  x         x  z         z  2{         4{  T|         V|  v}         x}  ~         ~             ܀         ހ                        "  B         D  d         f                        ʈ         ̈                        0                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               R!         f!  2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    ]
         ^
             E      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     K
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -;-    ;-:-    :-I-    I-|-d    |--    --0    --    -b-H	    b-c-P
    c-y-
    y-3׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               I!         V!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P           /         0  B         C  U	         V	  h
         i
  {         &  C         D  a         b                                                                                  @  2          a  t                                     !          E!  X"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q   w  operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       \                        =         @  ]         `  }                                                      	  
          
  =         @  ]         `  }                                                                    =         @  ]         `  }                                                                    =         @  ]         `  }                        !         !  "         "  #          $  %          %  :&         @&  Z'         `'  z(         (  )         )  *         *  +         +  ,          -  .          .  :/         @/  Z0         `0  z1          1  |2          2  3          3  4          4  5           6  6          7  :8         @8  Z9         `9  z:         :  ;         ;  <         <  =         =  >          ?  @          @  :A         G  I         I  (J         +J  HK         KK  hL         kL  M         M  N         N  O         O  P         P  R         R  (S         +S  HT         KT  hU         kU  V         V  W         W  X         X  Y         Y  [         [  (\         +\  H]         K]  h^         k^  _         _  `         `  a         a  b         b  d         d  4e         7e  Vf         Yf  xg         {g  h                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               @!         T!  1         1  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -/-    /-m<-    m<-K-    K-U~-d    U~--    --p    -b-    b--
    -)-
    )--t    -3׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               D!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        9         :  V         W  s         t                                      2          S  i                                              3  K          l                                     N          O   }!         ~!  "         "  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         :  A          `  g                                             $  9%         :%  Z&         \&  |'         ~'  (         (  )         )  *         *  ,         ,  &-         (-  H.         J.  j/         l/  0         0  1         1  2         2  3         3  5         5  66         86  X7         Z7  z8         |8  9         9  :         :  ;         ;  =         =  $>         &>  F?         H?  h@         j@  A         A  B         B  C         C  D         D  F         F  4G         6G  WH         XH  yI         zI  J         J  K         K  L         L  #N         $N  VO         WO  P         P  Q         Q  R         R  "T         #T  UU         VU  V         V  W         W  X         X  &Z         (Z  \[         ^[  \         \  ]         ]  ^          _  4`         6`  ja         la  b         b  c         c  
e         e  >f         @f  rg         tg  h         h  i         i  k         k  Bl         Dl  vm         xm  n         n  o         o  q         q  #r         $r  Es         Fs  gt         ht  u         u  v         v  w         w  x         x  z         z  3{         4{  U|         V|  w}         x}  ~         ~             ݀         ހ              !         "  C         D  e         f                        ˈ         ̈                        1                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               P!         f!  2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    ^
         _
             F      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     L
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     ---`     ---     -e-    e-1<-    1<-J-    J-~-d    ~--    -%-p    %-b-    b--
    --
    -簳-t    簳-?׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               _!         `!  2         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         ?         @  _         `                                               %  ,          ^  e                                              >  S          u                                     ;          <   Y!         Z!  w"         x"  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       %          D  K          j  q                     $  8%         :%  [&         \&  }'         ~'  (         (  )         )  *         *  ,         ,  '-         (-  I.         J.  k/         l/  0         0  1         1  2         2  3         3  5         5  76         86  Y7         Z7  {8         |8  9         9  :         :  ;         ;  =         =  %>         &>  G?         H?  i@         j@  A         A  B         B  C         C  D         D  F         F  5G         6G  VH         XH  xI         zI  J         J  K         K  L         L  N         $N  EO         WO  xP         P  Q         Q  R         R  T         #T  DU         VU  wV         V  W         W  X         X  Z         (Z  M[         ^[  \         \  ]         ]  ^          _  %`         6`  [a         la  b         b  c         c  d         e  /f         @f  cg         tg  h         h  i         i  j         k  3l         Dl  gm         xm  n         n  o         o   q         q  "r         $r  Ds         Fs  ft         ht  u         u  v         v  w         w  x         x  z         z  2{         4{  T|         V|  v}         x}  ~         ~             ܀         ހ                        "  B         D  d         f                        ʈ         ̈                        0                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               `!         j!  2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    Z
         [
             "      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     =
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     -G-    G-4-    4-C-    C-v-l    v-5-    5--x    -b-	    b--
    --
    -s-|    s-ճ-    ճ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               Q!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        9  M          r                                                >  5          _  V	          	  w
          
                                               =         >  \                      '         (  :         ;  M         N  `                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |        8"  Y#         [#  |$         ~$  %         %  &         &  '         '  )         )  0*         2*  T+         V+  x,         z,  -         -  .         .  /         /  1         
1  ,2         .2  P3         R3  t4         v4  5         5  6         6  7         7  9         9  (:         *:  L;         N;  p<         r<  =         =  >         >  ?         ?   A         A  $B         &B  HC         JC  lD         nD  E         E  F         F  G          `  %a          Ga  [b          }b  c          c  d          d  e         -l  Nm         Pm  qn         sn  o         o  p         p  q         q  s         s  %t         't  Iu         Ku  mv         ov  w         w  x         x  y         y  z         z  !|         #|  E}         G}  i~         k~                        Ձ         ׁ                        A         C  e         g                        щ         Ӊ           ړ                       "  D         F  h         j                        ԛ         ֛                        @         B  d         f                        У         ң                        <         >  `         b                        ̫         Ϋ                        8         :  \         ^                        ȳ         ʳ                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               P!         T!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    Z
         [
                   
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    Z
         [
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     k
            4     ---`     ---     -Y-    Y-%<-    %<-J-    J-~-d    ~--    -%-p    %-b-    b--
    --
    -簳-t    簳-?׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               [!         \!  	2         
2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        =         >  \         ]  {         |                                       (          Y  a                                              9  O          p                                    4          5   Q!         R!  n"         o"  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                   	           (   /!          N!  U"         $  9%         :%  Z&         \&  |'         ~'  (         (  )         )  *         *  ,         ,  &-         (-  H.         J.  j/         l/  0         0  1         1  2         2  3         3  5         5  66         86  X7         Z7  z8         |8  9         9  :         :  ;         ;  =         =  $>         &>  F?         H?  h@         j@  A         A  B         B  C         C  D         D  F         F  4G         6G  WH         XH  yI         zI  J         J  K         K  L         L  N         $N  FO         WO  yP         P  Q         Q  R         R  T         #T  EU         VU  xV         V  W         W  X         X  Z         (Z  L[         ^[  \         \  ]         ]  ^          _  $`         6`  Za         la  b         b  c         c  d         e  .f         @f  bg         tg  h         h  i         i  j         k  2l         Dl  fm         xm  n         n  o         o  q         q  #r         $r  Es         Fs  gt         ht  u         u  v         v  w         w  x         x  z         z  3{         4{  U|         V|  w}         x}  ~         ~             ݀         ހ              !         "  C         D  e         f                        ˈ         ̈                        1                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ^!         j!  2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    Z
         [
             "      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     >
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     -G-    G-4-    4-C-    C-v-l    v-5-    5--x    -b-	    b--
    --
    -s-|    s-ճ-    ճ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               Q!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        9  N          r                                                >  4          _  U	          	  v
          
                                   !         "  A         B  a                      ,         -  ?         @  R         S  e                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |        8"  Z#         [#  }$         ~$  %         %  &         &  '         '  )         )  1*         2*  U+         V+  y,         z,  -         -  .         .  /         /  	1         
1  -2         .2  Q3         R3  u4         v4  5         5  6         6  7         7  9         9  ):         *:  M;         N;  q<         r<  =         =  >         >  ?         ?  A         A  %B         &B  IC         JC  mD         nD  E         E  F         F  G          f  3g          Ug  ih          h  i          i  j          j  l         -l  Om         Pm  rn         sn  o         o  p         p  q         q  s         s  &t         't  Ju         Ku  nv         ov  w         w  x         x  y         y  z         z  "|         #|  F}         G}  j~         k~                        ց         ׁ                        B         C  f         g                        ҉         Ӊ           ړ             !         "  E         F  i         j                        ՛         ֛                        A         B  e         f                        ѣ         ң                        =         >  a         b                        ͫ         Ϋ                        9         :  ]         ^                        ɳ         ʳ                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               N!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    Z
         [
                    
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    Z
         [
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     l
            0     -,-X     ,--     -M-    M-4-x    4-C-|    C-v-\    v--    --h    -b-    b--n-	    -n--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               W!         X!  2         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        7  L          n                                                6  2          Y  U	          |	  x
          
                                              9         :  W                      ,         -  ?         @  R         S  e                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            "         $  F         H  j         l                                              	          	  B
         D
  f         h                                                         >         @  b         d                                                         :         <  ^         `                                              !         !  6"          T#  u$          $  %          %  &          '  >(          `(  )         .I  PJ         RJ  tK         vK  L         L  M         M  N         N  P         P  (Q         *Q  LR         NR  pS         rS  T         T  U         U  V         V   X         X  $Y         &Y  HZ         JZ  l[         n[  \         \  ]         ]  ^         ^  _         _  #a         $a  Ib         Jb  oc         pc  d         d  e         e  f         f  h         h  -i         .i  Sj         Tj  yk         m  n         n  p         p  <q         =q  `r         ar  s         s  t         t  u         u  v         v  x         x  8y         9y  \z         ]z  {         {  |         |  }         }  ~         ~             4         5  X         Y  |         }             ą         Ņ                        0         1  T         U  x         y                                                      >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               U!         X!  2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     P
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -S-    S-:-x    :-I-|    I-|-\    |-+-    +--h    -b-    b-x-	    x--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               T!         \!  2         
2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P           %         &  8         9  K	         L	  ^
         _
  q         !  >         ?  \         ]  z         {                                                              "          C  E          f  {                                      !          B!  W"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                     "  =         D  _         f                                              		         	  +
         2
  M         T  o         v                                                          ;         B  ]         d                                                         )         0  K         R  m         t                                     !         "  .#         3#  P$          ~$  %          %  &          &  '          '  (          )  *         sG  H         H  I         I  J         J  K         K  M         M  :N         ?N  \O         aO  ~P         P  Q         Q  R         R  S         S  U         U  (V         -V  JW         OW  lX         qX  Y         Y  Z         Z  [         [  \         \  ^         ^  8_         =_  Z`         _`  |a         a  b         b  c         c  d         d  f         	f  &g         +g  Hh         Mh  ji         oi  j         j  k         k  l         $  A         F  c         h                        ɕ         Ε                        /         4  Q         V  s         x                        ٞ         ޞ                       "  ?         D  a         f                        Ǧ         ̦                        -         2  O         T  q                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               M!         Z!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -G-    G-4-x    4-C-|    C-v-\    v--    --h    -b-    b--n-	    -n--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               S!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        7  M          n                                                6  1          Y  T	          |	  w
          
                                              5         6  R                      '         (  :         ;  M         N  `                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            #         $  G         H  k         l                                              	          	  C
         D
  g         h                                                         ?         @  c         d                                                         ;         <  _         `                                              !         !  7"          )  *          *  ,          ),  J-          l-  .          .  /         .I  QJ         RJ  uK         vK  L         L  M         M  N         N  P         P  )Q         *Q  MR         NR  qS         rS  T         T  U         U  V         V  X         X  %Y         &Y  IZ         JZ  m[         n[  \         \  ]         ]  ^         ^  _         _  "a         $a  Hb         Jb  nc         pc  d         d  e         e  f         f  h         h  ,i         .i  Rj         Tj  xk         m  n         n  p         p  ;q         =q  _r         ar  s         s  t         t  u         u  v         v  x         x  7y         9y  [z         ]z  {         {  |         |  }         }  ~         ~             3         5  W         Y  {         }             Å         Ņ                        /         1  S         U  w         y                                                      >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               Q!         T!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     O
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -M-    M-:-x    :-I-|    I-|-\    |-+-    +--h    -b-    b-x-	    x--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               R!         X!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P           %         &  8         9  K	         L	  ^
         _
  q         !  =         >  Z         [  w         x                                                                       >  ?          a  w                                     !          =!  S"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                     "  <         D  ^         f                                              	         	  *
         2
  L         T  n         v                                                          :         B  \         d  ~                                                       (         0  J         R  l         t                                     !         "  /#         3#  Q$          2*  4+          V+  X,          z,  |-          -  .          .  /         sG  H         H  I         I  J         J  K         K  M         M  ;N         ?N  ]O         aO  P         P  Q         Q  R         R  S         S  U         U  )V         -V  KW         OW  mX         qX  Y         Y  Z         Z  [         [  \         \  ^         ^  9_         =_  [`         _`  }a         a  b         b  c         c  d         d  f         	f  'g         +g  Ih         Mh  ki         oi  j         j  k         k  l         $  B         F  d         h                        ʕ         Ε                        0         4  R         V  t         x                        ڞ         ޞ                       "  @         D  b         f                        Ȧ         ̦                        .         2  P         T  r                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               M!         Z!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -G-    G-;-x    ;-I-|    I-|-\    |-+-    +--h    -b-    b-x-	    x--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               Q!         X!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P           /         0  B         C  U	         V	  h
         i
  {         !  E         F  j         k                                                                      =  @          ^  a                                               *   D!          c!  }"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                      "  B         D  d         f                                              	         	  0
         2
  R         T  t         v                                                          @         B  b         d                                                         .         0  P         R  r         t                                     !         "  1#         3#  S$          /  0          
1  2          .2  33          R3  W4          v4  {5         sG  H         H  I         I  J         J  K         K  M         M  =N         ?N  _O         aO  P         P  Q         Q  R         R  S         S  	U         U  +V         -V  MW         OW  oX         qX  Y         Y  Z         Z  [         [  \         \  ^         ^  ;_         =_  ]`         _`  a         a  b         b  c         c  d         d  f         	f  )g         +g  Kh         Mh  mi         oi  j         j  k         k  l         $  D         F  f         h                        ̕         Ε                        2         4  T         V  v         x                        ܞ         ޞ                        "  B         D  d         f                        ʦ         ̦                        0         2  R         T  t                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               R!         ^!  2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 }
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -M-    M-4-x    4-C-|    C-v-\    v--    --h    -b-    b-x-	    x--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               T!         \!  2         
2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        9  M          r                                                >  5          _  V	          	  w
          
                                               =         >  \                      '         (  :         ;  M         N  `                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                             "  '          L  Q         "  
$         $  '%         ,%  G&         L&  g'         l'  (         (  )         )  *         *  +         +  -         -  '.         ,.  G/         L/  g0         l0  1         1  2         2  3         3  4         4  6         6  '7         ,7  G8         L8  g9         l9  :         :  ;         ;  <         <  =         =  ?         ?  '@         ,@  GA         LA  gB         lB  C         C  D         D  E         AF  \G         aG  |H         H  I         I  J         J  K         K  L         M  N         !N  <O         AO  \P         aP  |Q         Q  R         R  S         S  T         T  U         V  W         !W  <X         AX  \Y         aY  |Z         Z  [         [  \         \  ]         ]  ^         _  `         !`  <a         Aa  \b         ab  |c         c  d         d  e         e  f         f  g         h  i         i  k         k  0l         2l  Pm         Rm  pn         rn  o         o  p         p  q         q  r         r  t         t  0u         2u  Pv         Rv  pw         rw  x         x  y         y  z         z  {         {  }         }  0~         2~  P         R  p         r                        Ѓ         ҃                        0         2  P         R  p                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               D!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -;-    ;-;-x    ;-I-|    I-|-\    |-+-    +--h    -b-    b-x-	    x--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               M!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P           /         0  B         C  U	         V	  h
         i
  {         !  F         G  l         m                                                           !  #          B  D          c  e                                               /   J!          h!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            !         "  C         D  e         f                                              	         	  1
         2
  S         T  u         v                                                          A         B  c         d                                                         /         0  Q         R  s         t                                     !         "  2#         3#  T$          5  6          6  7          7  8          9  :          *:  /;         sG  H         H  I         I  J         J  K         K  M         M  >N         ?N  `O         aO  P         P  Q         Q  R         R  S         S  
U         U  ,V         -V  NW         OW  pX         qX  Y         Y  Z         Z  [         [  \         \  ^         ^  <_         =_  ^`         _`  a         a  b         b  c         c  d         d  f         	f  *g         +g  Lh         Mh  ni         oi  j         j  k         k  l         $  E         F  g         h                        ͕         Ε                        3         4  U         V  w         x                        ݞ         ޞ              !         "  C         D  e         f                        ˦         ̦                        1         2  S         T  u                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               P!         ^!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 }
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -G-    G-4-x    4-C-|    C-v-\    v--    --h    -b-    b-x-	    x--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               R!         X!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        9  N          r                                                >  4          _  U	          	  v
          
                                   !         "  A         B  a                      ,         -  ?         @  R         S  e                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         v  {                                                #         "  $         $  (%         ,%  H&         L&  h'         l'  (         (  )         )  *         *  +         +  -         -  (.         ,.  H/         L/  h0         l0  1         1  2         2  3         3  4         4  6         6  (7         ,7  H8         L8  h9         l9  :         :  ;         ;  <         <  =         =  ?         ?  (@         ,@  HA         LA  hB         lB  C         C  D         D  E         AF  ]G         aG  }H         H  I         I  J         J  K         K  L         M  N         !N  =O         AO  ]P         aP  }Q         Q  R         R  S         S  T         T  U         V  W         !W  =X         AX  ]Y         aY  }Z         Z  [         [  \         \  ]         ]  ^         _  `         !`  =a         Aa  ]b         ab  }c         c  d         d  e         e  f         f  g         h  i         i  k         k  1l         2l  Qm         Rm  qn         rn  o         o  p         p  q         q  r         r  t         t  1u         2u  Qv         Rv  qw         rw  x         x  y         y  z         z  {         {  }         }  1~         2~  Q         R  q         r                        у         ҃                        1         2  Q         R  q                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               F!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -A-    A-4-x    4-C-|    C-v-\    v--    --h    -b-    b-x-	    x--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               Q!         X!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        7  L          n                                                6  2          Y  U	          |	  x
          
                                              9         :  W                      ,         -  ?         @  R         S  e                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         H  P          r  z                                             "  $         $  *%         ,%  J&         L&  j'         l'  (         (  )         )  *         *  +         +  
-         -  *.         ,.  J/         L/  j0         l0  1         1  2         2  3         3  4         4  
6         6  *7         ,7  J8         L8  j9         l9  :         :  ;         ;  <         <  =         =  
?         ?  *@         ,@  JA         LA  jB         lB  C         C  D         D  E         AF  _G         aG  H         H  I         I  J         J  K         K  L         M  N         !N  ?O         AO  _P         aP  Q         Q  R         R  S         S  T         T  U         V  W         !W  ?X         AX  _Y         aY  Z         Z  [         [  \         \  ]         ]  ^         _  `         !`  ?a         Aa  _b         ab  c         c  d         d  e         e  f         f  g         h  i         i  k         k  1l         2l  Qm         Rm  qn         rn  o         o  p         p  q         q  r         r  t         t  1u         2u  Qv         Rv  qw         rw  x         x  y         y  z         z  {         {  }         }  1~         2~  Q         R  q         r                        у         ҃                        1         2  Q         R  q                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               I!         T!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 }
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-:-x    :-I-|    I-|-\    |--    --h    -]b-    ]b---	    ---h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               F!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P           %         &  8         9  K	         L	  ^
         _
  q         !  >         ?  \         ]  z         {                                                              "          C  E          f  {                                      !          B!  W"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       5         <  S         Z  q         x                                              
         
  %         ,  C         J  a         h                                                                    3         :  Q         X  o         v                                                         #         *  A         H  _          f   }!         p"  #         #  $         $  %         %  &         &  (         (  )         $)  =*         B*  [+         `+  y,         ~,  -         -  .         .  /         /  0         0  2         2  -3         23  K4         P4  i5         n5  6         6  7         7  8         8  9         9  :         ;  <         "<  ;=         @=  Y>         ^>  w?         |?  @         @  A         A  B         B  C         C  E         E  +F         0F  IG         :H  SI         XI  qJ         vJ  K         K  L         L  M         M  N         N  P         P  %Q         *Q  CR         HR  aS         fS  T         T  U         U  V         V  W         W  X         X  Z         Z  3[         8[  Q\         V\  o]         t]  ^         ^  _         _  `         `  a         a  c         
c  #d         (d  Ae         Fe  _f                        ʄ                        .          P  `                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               -!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   [      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -;-    ;-4-x    4-C-|    C-v-\    v--    --h    -b-    b-x-	    x--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               M!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        7  M          n                                                6  1          Y  T	          |	  w
          
                                              5         6  R                      '         (  :         ;  M         N  `                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |           "          D  L          n  v              !          !  "         "  
$         $  +%         ,%  K&         L&  k'         l'  (         (  )         )  *         *  +         +  -         -  +.         ,.  K/         L/  k0         l0  1         1  2         2  3         3  4         4  6         6  +7         ,7  K8         L8  k9         l9  :         :  ;         ;  <         <  =         =  ?         ?  +@         ,@  KA         LA  kB         lB  C         C  D         D  E         AF  `G         aG  H         H  I         I  J         J  K         K   M         M   N         !N  @O         AO  `P         aP  Q         Q  R         R  S         S  T         T   V         V   W         !W  @X         AX  `Y         aY  Z         Z  [         [  \         \  ]         ]   _         _   `         !`  @a         Aa  `b         ab  c         c  d         d  e         e  f         f   h         h  i         i  k         k  0l         2l  Pm         Rm  pn         rn  o         o  p         p  q         q  r         r  t         t  0u         2u  Pv         Rv  pw         rw  x         x  y         y  z         z  {         {  }         }  0~         2~  P         R  p         r                        Ѓ         ҃                        0         2  P         R  p                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               E!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 }
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-     5-:-     :-{I-     {I-|-     |--2     --
     -cb-:     cb---b     ---R     -3׳-*                    =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               H!         T!  1         1  =                              b              )      operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81           w         5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                     b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                                 #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P           %         &  8         9  K	         L	  ^
         _
  q         !  =         >  Z         [  w         x                                                                       >  ?          a  w                                     !          =!  S"                                    b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                            b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                                 3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       4         <  R         Z  p         x                                              
         
  $         ,  B         J  `         h  ~                                                                  2         :  P         X  n         v                                                         "         *  @         H  ^          f   |!         p"  #         #  $         $  %         %  &         &  (         (   )         $)  >*         B*  \+         `+  z,         ~,  -         -  .         .  /         /  0         0  2         2  .3         23  L4         P4  j5         n5  6         6  7         7  8         8  9         9   ;         ;  <         "<  <=         @=  Z>         ^>  x?         |?  @         @  A         A  B         B  C         C  E         E  ,F         0F  JG         :H  TI         XI  rJ         vJ  K         K  L         L  M         M  N         N  P         P  &Q         *Q  DR         HR  bS         fS  T         T  U         U  V         V  W         W  X         X  Z         Z  4[         8[  R\         V\  p]         t]  ^         ^  _         _  `         `  a         a  c         
c  $d         (d  Be         Fe  `f                        Ċ                        (          J  Z                         >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               /!         D!  1         1  =                                    b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1                    	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   \            
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                           b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-;-x    ;-I-|    I-|-\    |--    --h    -cb-    cb-3-	    3--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               C!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P           /         0  B         C  U	         V	  h
         i
  {         !  E         F  j         k                                                                      =  @          ^  a                                               *   D!          c!  }"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       :         <  X         Z  v         x                                              
         
  *         ,  H         J  f         h                                                                    8         :  V         X  t         v                                              
           (         *  F         H  d          f   !         p"  #         #  $         $  %         %  &         &  (         (  ")         $)  @*         B*  ^+         `+  |,         ~,  -         -  .         .  /         /  0         0  2         2  03         23  N4         P4  l5         n5  6         6  7         7  8         8  9         9  ;         ;   <         "<  >=         @=  \>         ^>  z?         |?  @         @  A         A  B         B  C         C  E         E  .F         0F  LG         :H  VI         XI  tJ         vJ  K         K  L         L  M         M  N         N  
P         P  (Q         *Q  FR         HR  dS         fS  T         T  U         U  V         V  W         W  X         X  Z         Z  6[         8[  T\         V\  r]         t]  ^         ^  _         _  `         `  a         a  c         
c  &d         (d  De         Fe  bf          l  n          6n  Zo          yo  p          p  q          q  #s                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               2!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   d      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-<-x    <-J-|    J-~-\    ~--    --h    -?b-    ?b---	    ---h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               D!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        ;         <  Y         Z  w         x                                    !  6          X  m                                              8  L          q                                    3          4   R!         S!  q"         r"  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       5         :  R         W  o         t                                               
         
           "  :         ?  W         \  t         y                                                       
  "         '  ?         D  \         a  y         ~                                              
           '         ,  D          I   a!         d#  ~$         $  %         %  &         &  '         '  )         )  '*         ,*  K+         P+  o,         t,  -         -  .         .  /         /  0         1  !2         &2  C3         H3  e4         j4  5         5  6         6  7         7  8         8  :         :  1;         6;  S<         X<  s=         x=  >         >  ?         ?  @         @  A         A  C         C  3D         8D  SE         XE  sF         xF  G          G  H          H  J          2J  QK          pK  L          L  M         m  n         n  9o         ;o  Up         Wp  qq         sq  r         r  s         s  t         t  u         u  v         v  x         x  5y         7y  Qz         Sz  m{         o{  |         |  }         }  ~         ~                                   1         3  M         O  i         k                                   ى         ۉ                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               $!         2!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   [      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-;-x    ;-I-|    I-|-\    |--    --h    -cb-    cb-3-	    3--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               C!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P           /         0  B         C  U	         V	  h
         i
  {         !  F         G  l         m                                                           !  #          B  D          c  e                                               /   J!          h!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       ;         <  Y         Z  w         x                                              
         
  +         ,  I         J  g         h                                                                    9         :  W         X  u         v                                                         )         *  G         H  e          f   !         p"  #         #  $         $  %         %  &         &  (         (  #)         $)  A*         B*  _+         `+  },         ~,  -         -  .         .  /         /  0         0  2         2  13         23  O4         P4  m5         n5  6         6  7         7  8         8  9         9  ;         ;  !<         "<  ?=         @=  ]>         ^>  {?         |?  @         @  A         A  B         B  C         C  E         E  /F         0F  MG         :H  WI         XI  uJ         vJ  K         K  L         L  M         M  N         N  P         P  )Q         *Q  GR         HR  eS         fS  T         T  U         U  V         V  W         W  X         X  Z         Z  7[         8[  U\         V\  s]         t]  ^         ^  _         _  `         `  a         a  	c         
c  'd         (d  Ee         Fe  cf          Bs  ft          t  u          u  v          w  /x          Nx  ry                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               2!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   e      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-<-x    <-J-|    J-~-\    ~--    --h    -Eb-    Eb---	    ---h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               B!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        9         :  V         W  s         t                                      2          S  i                                              3  H          l                                    0          1   P!         Q!  p"         q"  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       6         :  S         W  p         t                                              
         
           "  ;         ?  X         \  u         y                                                       
  #         '  @         D  ]         a  z         ~                                                         (         ,  E          I   b!         d#  $         $  %         %  &         &  '         '  )         )  (*         ,*  L+         P+  p,         t,  -         -  .         .  /         /   1         1  "2         &2  D3         H3  f4         j4  5         5  6         6  7         7  8         8  :         :  2;         6;  T<         X<  t=         x=  >         >  ?         ?  @         @  A         A  C         C  4D         8D  TE         XE  tF         xF  G          M  O          *O  IP          hP  Q          Q  R          R  T         m  n         n  :o         ;o  Vp         Wp  rq         sq  r         r  s         s  t         t  u         u  v         v  x         x  6y         7y  Rz         Sz  n{         o{  |         |  }         }  ~         ~                                   2         3  N         O  j         k                                   ډ         ۉ                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               (!         6!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   \      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-7<-x    7<-J-|    J-~-\    ~--    --h    -Eb-    Eb---	    ---h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               I!         V!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             $         %  I         J  n         o                                               >  X          w                                       5          W  l                                   6         7  T          U   r!         s!  "         "  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       8         :  U         W  r         t                                              
         
            "  =         ?  Z         \  w         y                                                       
  %         '  B         D  _         a  |         ~                                                         *         ,  G          I   d!         d#  $         $  %         %  &         &  '         '  )         )  **         ,*  N+         P+  r,         t,  -         -  .         .  /         /  1         1  $2         &2  F3         H3  h4         j4  5         5  6         6  7         7  8         8  :         :  4;         6;  V<         X<  v=         x=  >         >  ?         ?  @         @  A         A  C         C  6D         8D  VE         XE  vF         xF  G          "T  DU          `U  V          V  W          W  X          Y  <Z         m  n         n  :o         ;o  Vp         Wp  rq         sq  r         r  s         s  t         t  u         u  v         v  x         x  6y         7y  Rz         Sz  n{         o{  |         |  }         }  ~         ~                                   2         3  N         O  j         k                                   ډ         ۉ                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               )!         6!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   d      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-y-     y--    -:-x    :-{I-|    {I-|-\    |--    --P    -ub-    ub-c-	    c-s-D
    s-'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ;!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P           %         &  8         9  K	         L	  ^
         _
  q         !  >         ?  \         ]  z         {                                                              "          C  E          f  {                                      !          B!  W"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p                     "  =         D  _         f                                              		         	  +
         2
  M         T  o         v                                                          ;         B  ]         d                                                         )         0  K         R  m         t                                     !         !  #         #  2$         7$  Q%         V%  p&         u&  '         '  (         (  )         )  +         +  6,         =,  ^-         e-  .         .  /         /  0         0  1         2  (3         /3  R4         Y4  |5         5  6         6  7         7  8         9  $:         +:  N;         U;  x<         <  =         =  >         >  ?         @  &A         -A  RB         YB  ~C         C  D         D  E         E  G         	G  .H         5H  OI          n  o         o  9p         >p  Xq         ]q  wr         |r  s         s  t         t  u         u  v         v  x         x  1y         6y  Pz         Uz  o{         t{  |         |  }         }  ~         ~             
           )         .  H         M  g         l                        ć                                  ߗ                        ;  J                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               =!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-7<-x    7<-J-|    J-~-\    ~--    --h    -Kb-    Kb-3-	    3--h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               E!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             %         &  K         L  q         r                                   
  %          C  ^          |                                    %  ;          \  r                                   :         ;  W          X   t!         u!  "         "  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       9         :  V         W  s         t                                              
         
  !         "  >         ?  [         \  x         y                                              	         
  &         '  C         D  `         a  }         ~                                                         +         ,  H          I   e!         d#  ~$         $  %         %  &         &  '         '  )         )  +*         ,*  O+         P+  s,         t,  -         -  .         .  /         /  1         1  %2         &2  G3         H3  i4         j4  5         5  6         6  7         7  8         8  :         :  5;         6;  W<         X<  w=         x=  >         >  ?         ?  @         @  A         A  C         C  7D         8D  WE         XE  wF         xF  G          XZ  z[          [  \          \  ]          ^  4_          P_  r`         m  n         n  9o         ;o  Up         Wp  qq         sq  r         r  s         s  t         t  u         u  v         v  x         x  5y         7y  Qz         Sz  m{         o{  |         |  }         }  ~         ~                                   1         3  M         O  i         k                                   ى         ۉ                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               +!         :!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   e      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-y-     y--    -:-x    :-uI-|    uI-|-\    |- -     --P    -ub-    ub-c-	    c-s-D
    s-'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               :!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P           %         &  8         9  K	         L	  ^
         _
  q         !  =         >  Z         [  w         x                                                                       >  ?          a  w                                     !          =!  S"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p                     "  <         D  ^         f                                              	         	  *
         2
  L         T  n         v                                                          :         B  \         d  ~                                                       (         0  J         R  l         t                                     !         !  #         #  3$         7$  R%         V%  q&         u&  '         '  (         (  )         )  +         +  5,         =,  ]-         e-  .         .  /         /  0         0  1         2  '3         /3  Q4         Y4  {5         5  6         6  7         7  8         9  #:         +:  M;         U;  w<         <  =         =  >         >  ?         @  %A         -A  QB         YB  }C         C  D         D  E         E  G         	G  -H         5H  PI          n  o         o  :p         >p  Yq         ]q  xr         |r  s         s  t         t  u         u  v         v  x         x  2y         6y  Qz         Uz  p{         t{  |         |  }         }  ~         ~                        *         .  I         M  h         l                        Ň          i  x                      ŝ  Ԟ                      !  0                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               =!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     --    -:-x    :-I-|    I-|-\    |-&-    &--P    -{b-    {b-c-	    c-y-D
    y-'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               B!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P           /         0  B         C  U	         V	  h
         i
  {         !  E         F  j         k                                                                      =  @          ^  a                                               *   D!          c!  }"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p                      "  B         D  d         f                                              	         	  0
         2
  R         T  t         v                                                          @         B  b         d                                                         .         0  P         R  r         t                                     !         !  #         #  5$         7$  T%         V%  s&         u&  '         '  (         (  )         )  +         +  ;,         =,  c-         e-  .         .  /         /  0         0  2         2  -3         /3  W4         Y4  5         5  6         6  7         7  8         9  ):         +:  S;         U;  }<         <  =         =  >         >  ?         @  +A         -A  WB         YB  C         C  D         D  E         E  G         	G  3H         5H  RI          n  o         o  <p         >p  [q         ]q  zr         |r  s         s  t         t  u         u  v         v  x         x  4y         6y  Sz         Uz  r{         t{  |         |  }         }  ~         ~                        ,         .  K         M  j         l                        Ǉ          O  a          }                        ٤                                 >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               F!         T!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,-m-     m--    -<-    <-J-    J-}-l    }-	-    	--P    --    -b-h	    b-傳-p
    傳--
    -۰-    ۰-'׳-L                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               6!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        ;         <  Y         Z  w         x                                    !  6          X  m                                              8  L          q                                    3          4   R!         S!  q"         r"  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        #  $         $  %         %  &         '  #(         *(  G)         N)  k*         r*  +         +  ,         ,  -         -  .         /  0         &0  C1         J1  g2         n2  3         3  4         4  5         5  6         6  8         "8  ?9         F9  c:         j:  ;         ;  <         <  =         =  >         >  @         @  ;A         BA  _B         fB  C         C  D         D  E         E  F         F  H         H  6I         ;I  WJ         \J  xK         }K  L         L  M         M  N         N  P         
P  ,Q         .Q  PR         RR  tS         vS  T         T  U         U  V         V  X         X  &Y         (Y  HZ         JZ  j[         l[  \         \  ]         ]  ^         ^  _         _  a         a  6b         8b  Xc         Zc  xd         zd  e         e  f         f  g         g  h         h  j         j  8k         :k  Xl         Zl  xm         zm  n         n  o         o  p         p  q         q  s         s  :t         ?t  [u         `u  |v         v  w         w  x         x  y         y   {         {  !|         &|  B}         G}  c~          h~  g                                  ˁ  ʂ                       )         .  J         O  k         p                                           b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               T!         ^!  2         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    Z
         [
                                           b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     K
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     --    -;-x    ;-I-|    I-|-\    |-g,-    g,--P    -{b-    {b-c-	    c-y-D
    y-'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               A!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P           /         0  B         C  U	         V	  h
         i
  {         !  F         G  l         m                                                           !  #          B  D          c  e                                               /   J!          h!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            !         "  C         D  e         f                                              	         	  1
         2
  S         T  u         v                                                          A         B  c         d                                                         /         0  Q         R  s         t                                     !         !  #         #  6$         7$  U%         V%  t&         u&  '         '  (         (  )         )  +         +  <,         =,  d-         e-  .         .  /         /  0         0  2         2  .3         /3  X4         Y4  5         5  6         6  7         7   9         9  *:         +:  T;         U;  ~<         <  =         =  >         >   @         @  ,A         -A  XB         YB  C         C  D         D  E         E  G         	G  4H         5H  SI          n  o         o  =p         >p  \q         ]q  {r         |r  s         s  t         t  u         u  v         v  x         x  5y         6y  Tz         Uz  s{         t{  |         |  }         }  ~         ~                        -         .  L         M  k         l                        ȇ          5  G          c  u                        ѫ                               >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               H!         T!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,-m-     m--    -<-    <-J-    J-}-l    }--    --P    --    -b-h	    b-傳-p
    傳--
    -ᰳ-    ᰳ-'׳-L                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               6!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        9         :  V         W  s         t                                      2          S  i                                              3  H          l                                    0          1   P!         Q!  p"         q"  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        #  $         $  %         %  &         '  "(         *(  F)         N)  j*         r*  +         +  ,         ,  -         -  .         /  0         &0  B1         J1  f2         n2  3         3  4         4  5         5  6         6  8         "8  >9         F9  b:         j:  ;         ;  <         <  =         =  >         >  @         @  :A         BA  ^B         fB  C         C  D         D  E         E  F         F  H         H  7I         ;I  XJ         \J  yK         }K  L         L  M         M  N         N  	P         
P  -Q         .Q  QR         RR  uS         vS  T         T  U         U  V         V  X         X  'Y         (Y  IZ         JZ  k[         l[  \         \  ]         ]  ^         ^  _         _  a         a  7b         8b  Yc         Zc  yd         zd  e         e  f         f  g         g  h         h  j         j  9k         :k  Yl         Zl  ym         zm  n         n  o         o  p         p  q         q  s         s  ;t         ?t  \u         `u  }v         v  w         w  x         x  y         y  {         {  "|         &|  C}         G}  d~         h~                        ǁ         ˁ             	                      .  -          O  N          p  o                                          b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               R!         ^!  2         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    Z
         [
                                            b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     L
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,-s-     s--    -1<-    1<-J-    J-~-d    ~-	-    	--p    -b-    b-傳-
    傳--
    -ᰳ-t    ᰳ-!׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               =!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             $         %  I         J  n         o                                               >  X          w                                       5          W  l                                   6         7  T          U   r!         s!  "         "  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       *  5          T  _          ~                       #  $         $  %         %  '         '  ((         *(  L)         N)  p*         r*  +         +  ,         ,  -         -   /         /  $0         &0  H1         J1  l2         n2  3         3  4         4  5         5  6         6   8         "8  D9         F9  h:         j:  ;         ;  <         <  =         =  >         >  @         @  @A         BA  dB         fB  C         C  D         D  E         E  F         F  H         H  9I         ;I  ZJ         \J  {K         }K  L         L  M         M  N         N  	P         
P  -Q         .Q  QR         RR  uS         vS  T         T  U         U  V         V  X         X  'Y         (Y  IZ         JZ  k[         l[  \         \  ]         ]  ^         ^  _         _  a         a  7b         8b  Yc         Zc  yd         zd  e         e  f         f  g         g  h         h  j         j  9k         :k  Yl         Zl  ym         zm  n         n  o         o  p         p  q         q  s         s  =t         ?t  ^u         `u  v         v  w         w  x         x  y         y  {         {  $|         &|  E}         G}  f~         h~                        Ɂ         ˁ                        ,         .  M         O  n         p                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ]!         b!  2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    Z
         [
             "      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     P
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                 w        ܯ--0     -y-     y-]-     ]-	B-P    	B-c-    c--    -!׳-p                             b      *      7   I      operation.0     sync             :            2            b         reason  ExchangeBarrier value   1       invocations     1                                     b      *      7   I      operation.0     sync             :            2            b         reason  ExchangeBarrier value   3       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   5       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   2       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,-s-     s--    -1<-    1<-J-    J-~-d    ~-	-    	--p    -b-    b-傳-
    傳--
    -ᰳ-t    ᰳ-!׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               =!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             %         &  K         L  q         r                                   
  %          C  ^          |                                    %  ;          \  r                                   :         ;  W          X   t!         u!  "         "  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                 &  1	          P	  [
          z
           #  $         $  %         %  '         '  )(         *(  M)         N)  q*         r*  +         +  ,         ,  -         -  /         /  %0         &0  I1         J1  m2         n2  3         3  4         4  5         5  6         6  !8         "8  E9         F9  i:         j:  ;         ;  <         <  =         =  >         >  @         @  AA         BA  eB         fB  C         C  D         D  E         E  F         F  H         H  :I         ;I  [J         \J  |K         }K  L         L  M         M  N         N  P         
P  ,Q         .Q  PR         RR  tS         vS  T         T  U         U  V         V  X         X  &Y         (Y  HZ         JZ  j[         l[  \         \  ]         ]  ^         ^  _         _  a         a  6b         8b  Xc         Zc  xd         zd  e         e  f         f  g         g  h         h  j         j  8k         :k  Xl         Zl  xm         zm  n         n  o         o  p         p  q         q  s         s  >t         ?t  _u         `u  v         v  w         w  x         x  y         y  {         {  %|         &|  F}         G}  g~         h~                        ʁ         ˁ                        -         .  N         O  o         p                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ]!         b!  2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    Z
         [
             "      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     Q
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                 w        Sܯ--0     -y-     y-]-     ]-	B-P    	B-c-    c--    -!׳-p                             b      *      7   I      operation.0     sync             :            2            b         reason  ExchangeBarrier value   1       invocations     1                                     b      *      7   I      operation.0     sync             :            2            b         reason  ExchangeBarrier value   3       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   5       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   2       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    <     -,-p     ,-C-     C--    -4-    4-C-    C-v-t    v-30-     30-7-X    7--    -ob-p	    ob-߂-x
    ߂--
    -U-    U-ճ-\    ճ-׳-<                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     p            !          J!  1          1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        5  F          j  {                                  	            .  2          S  W	          x	  |
          
                                              1         2  M                      0         1  E         F  Z         [  o                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        T$  v%         x%  &         vL  M         M  N         N  O         O  Q         Q  )R         *R  MS         NS  qT         rT  U         U  V         V  W         W  Y         Y  %Z         &Z  I[         J[  m\         n\  ]         ]  ^         ^  _         _  `         `  !b         "b  Ec         Fc  id         jd  e         e  f         f  g         g  h         h  j         j  Ak         Bk  el         fl  m         m  n         n  o         o  p         p  r         r  =s         >s  at         bt  u         u  v         v  w         w  x         x  z         z  9{         :{  ]|         ^|  }         }  ~         ~                                   5         6  Y         Z  }         ~             Ň         Ƈ                        1         2  U         V  y         z                                            
  ,         .  P         R  t         v                                               	          *  -          N  Q          r  u                       ܠ         ޠ              $         &  H         J  l         n                        ب         ڨ                       "  D         F  h         j                        ԰                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               K!         L!  1         1  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    \
         ]
             $                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    \
         ]
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,-s-     s--    -1<-    1<-J-    J-~-d    ~-[-    [--p    -b-    b-낳-
    낳--
    -ϰ-t    ϰ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     }            -!          `!  1          2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        7         8  S         T  o         p                                      .          R  c                                              2  N          m                                    =          >   f!         g!  "         "  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                     3  9          Z  `                                 $  %         %  &         &  (         (  ')         ()  M*         N*  s+         t+  ,         ,  -         -  .         .  0         0  11         21  W2         X2  }3         ~3  4         4  5         5  6         6  8         8  ;9         <9  a:         b:  ;         ;  <         <  =         =  >         >  @          @  EA         FA  kB         lB  C         C  D         D  E         E  G         G  (H         *H  NI         PI  tJ         vJ  K         K  L         L  M         M  O         O  2P         4P  XQ         ZQ  ~R         R  S         S  T         T  U         U  W         W  <X         >X  bY         dY  Z         Z  [         [  \         \  ]         ]   _         "_  F`         H`  la         na  b         b  c         c  d         d  f         
f  0g         2g  Xh         Zh  i         i  j         j  k         k  l         l   n         "n  Ho         Jo  np         pp  q         q  r         r  s         s  u         u  ,v         .v  Rw         Tw  xx         zx  y         y  z         z  {         {  }         }  6~         8~  \         ^                        ΂         Ђ                        @         B  f         h                        ؊         ڊ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               i!         j!  2         2  >                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    \
         ]
             0      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     G
                   
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            <     -,-p     ,-I-     I--    -4-    4-{C-    {C-v-t    v-30-     30-7-X    7--    -ob-p	    ob-߂-x
    ߂--
    -U-    U-ճ-\    ճ-׳-<                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     q            !          N!  1          1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        5  G          j  |                                  	            .  1          S  V	          x	  {
          
                                              -         .  H                      +         ,  @         A  U         V  j                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        T$  w%         x%  &         vL  M         M  N         N  O         O  Q         Q  (R         *R  LS         NS  pT         rT  U         U  V         V  W         W   Y         Y  $Z         &Z  H[         J[  l\         n\  ]         ]  ^         ^  _         _  `         `   b         "b  Dc         Fc  hd         jd  e         e  f         f  g         g  h         h  j         j  @k         Bk  dl         fl  m         m  n         n  o         o  p         p  r         r  <s         >s  `t         bt  u         u  v         v  w         w  x         x  z         z  8{         :{  \|         ^|  }         }  ~         ~                                   4         6  X         Z  |         ~             ć         Ƈ                        0         2  T         V  x         z                                   	         
  -         .  Q         R  u         v                                              )         *  M         N  q         r                                   ޠ                        &  )          J  M         n                        ٨         ڨ             !         "  E         F  i         j                        հ                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               K!         L!  1         1  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    \
         ]
             $                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    \
         ]
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,-g-     g--    -1<-    1<-J-    J-~-d    ~-[-    [--p    -b-    b--
    --
    -ϰ-t    ϰ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     |            *!          \!  1          
2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        5         6  P         Q  k         l                                      *          M  _                                              -  J          h                                    :          ;   d!         e!  "         "  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                   #           D   J!          k!  q"         $  %         %  &         &   (         (  &)         ()  L*         N*  r+         t+  ,         ,  -         -  .         .  
0         0  01         21  V2         X2  |3         ~3  4         4  5         5  6         6  8         8  :9         <9  `:         b:  ;         ;  <         <  =         =  >         >  @          @  DA         FA  jB         lB  C         C  D         D  E         E  G         G  )H         *H  OI         PI  uJ         vJ  K         K  L         L  M         M  O         O  3P         4P  YQ         ZQ  R         R  S         S  T         T  U         U  W         W  =X         >X  cY         dY  Z         Z  [         [  \         \  ]         ]  !_         "_  G`         H`  ma         na  b         b  c         c  d         d  	f         
f  1g         2g  Yh         Zh  i         i  j         j  k         k  l         l  !n         "n  Io         Jo  op         pp  q         q  r         r  s         s  u         u  -v         .v  Sw         Tw  yx         zx  y         y  z         z  {         {  }         }  7~         8~  ]         ^                        ς         Ђ                        A         B  g         h                        ي         ڊ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               i!         j!  2         2  >                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    \
         ]
             2      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     H
                   
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,-I-     I--    -<-x    <-J-|    J-}-\    }-]3-    ]3--h    -{b-    {b-n-	    n--h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     p            !          L!  1          1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        =         >  \         ]  {         |                                      ,          U  g                                               5  F          j  {                                  2          3   N!         O!  j"         k"  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                      '  G         N  n         u                                   
           1	         8	  X
         _
                                                       "  B         I  i         p                                              ,         3  S         Z  z                                                       =          D   d!         k!  "          (<  ;=          \=  o>          >  ?          ?  @          @  B         K  L         L  N         N  5O         >O  cP         lP  Q         Q  R         R  S         S  U         $U  IV         RV  wW         W  X         X  Y         Y  Z         [  '\         0\  S]         \]  ^         ^  _         _  `         `  b         b  /c         8c  Yd         bd  e         e  f         f  g         g  i         
i  +j         4j  Uk         ^k  l         l  m         m  n         ȑ                        0         4  T         X  x         |                      ę                        ,         0  P         T  t         x                                              (         ,  L         P  p         t                        ܪ                       $         (  H         L  l         p                        ز         ܲ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               N!         R!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     D
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,--     -#-    #-5-x    5-C-|    C-v-\    v--    --h    -]b-    ]b-n-	    n--h
    -׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               F!         X!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        ;  W          v                                    '  *          F  I          e  h	          	  
          
                                   <         =  e         f           *  >         ?  S         T  h         i  }         ~                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            #         %  H         J  m         o                                              &	         (	  K
         M
  p         r                                              )         +  N         P  s         u                                            	  ,         .  Q         S  v         x                                   
             /!         1!  T"          T#  x$          $  %          %  &          '  A(          `(  )         I  J         J  K         K  M         M  &N         (N  KO         MO  pP         rP  Q         Q  R         R  S         S  U         U  )V         +V  NW         PW  sX         uX  Y         Y  Z         Z  [         [  ]         	]  ,^         .^  Q_         S_  v`         x`  a         a  b         b  c         c  e         e  4f         6f  Zg         \g  h         h  i         i  j         j  k         k  m         m  Bn         Dn  jo         lo  p         p  q         q  r         r  
t         t  2u         4u  Zv         \v  w         w  x         x  y         y  z         z  "|         $|  J}         L}  r~         t~                      Ā                        :         <  b         d                        ڈ         ܈             *         ,  R         T  z         |                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     }            )!          X!  1          2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     O
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-I-     I--    -<-x    <-J-|    J-}-\    }-]3-    ]3--h    -b-    b-n-	    n--h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     q            !          L!  1          1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         ?         @  _         `                                                 2          Z  m                                              :  L          o                                     5          6   P!         Q!  k"         l"  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            !         '  H         N  o         u                                              2	         8	  Y
         _
                                                       "  C         I  j         p                                              -         3  T         Z  {                                                       >          D   e!         k!  "          ,B  ?C          `C  sD          D  E          E  F          F  H         K  L         L  N         N  4O         >O  bP         lP  Q         Q  R         R  S         S  U         $U  HV         RV  vW         W  X         X  Y         Y  Z         [  &\         0\  R]         \]  ~^         ^  _         _  `         `  b         b  .c         8c  Xd         bd  e         e  f         f  g         g   i         
i  *j         4j  Tk         ^k  ~l         l  m         m  n         ȑ                        1         4  U         X  y         |                      ę             	           -         0  Q         T  u         x                                              )         ,  M         P  q         t                        ݪ                      %         (  I         L  m         p                        ٲ         ܲ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               P!         V!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     C
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,--     --    -5-x    5-C-|    C-v-\    v--    --h    -Wb-    Wb-n-	    n--h
    -׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               B!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        ;  X          v                          	          '  )          F  H          e  g	          	  
          
                                   ?         @  i         j           /  C         D  X         Y  m         n                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            $         %  I         J  n         o                                              '	         (	  L
         M
  q         r                                              *         +  O         P  t         u                                            	  -         .  R         S  w         x                                                0!         1!  U"          )  *          *  
,          ),  M-          l-  .          .  /         I  J         J  K         K  M         M  'N         (N  LO         MO  qP         rP  Q         Q  R         R  S         S  U         U  *V         +V  OW         PW  tX         uX  Y         Y  Z         Z  [         [  ]         	]  -^         .^  R_         S_  w`         x`  a         a  b         b  c         c  e         e  5f         6f  [g         \g  h         h  i         i  j         j  k         k  m         m  Cn         Dn  ko         lo  p         p  q         q  r         r  t         t  3u         4u  [v         \v  w         w  x         x  y         y  z         z  #|         $|  K}         L}  s~         t~             À         Ā                        ;         <  c         d                        ۈ         ܈             +         ,  S         T  {         |                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     |            &!          T!  1          1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     N
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-g-     g--    -4-x    4-C-|    C-v-\    v--    --h    -9b-    9b-	n-	    	n--h
    -!׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               3!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        5  F          j  {                                  	            .  2          S  W	          x	  |
          
                                              1         2  M                      0         1  E         F  Z         [  o                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                     %  C         J  h         o                                              !	         (	  F
         M
  k         r                                              $         +  I         P  n         u                                            	  '         .  L         S  q         x                                                *!         1!  O"          T#  l$          $  %          %  &          '   (          D(  \)         I  J         J  K         K  L         M  !N         (N  FO         MO  kP         rP  Q         Q  R         R  S         S  T         U  $V         +V  IW         PW  nX         uX  Y         Y  Z         Z  [         [  ]         	]  '^         .^  L_         S_  q`         x`  a         a  b         b  c         c  	e         e  /f         6f  Ug         \g  {h         h  i         i  j         j  k         k  m         m  ;n         Dn  co         lo  p         p  q         q  r         r  t         t  +u         4u  Sv         \v  {w         w  x         x  y         y  z         z  |         $|  C}         L}  k~         t~                      Ā                        3         <  [         d                        ӈ         ܈             #         ,  K         T  s         |                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     p            !          H!  1          1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     D
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-:-x    :-oI-|    oI-|-\    |--    --h    -ib-    ib-x-	    x--h
    -!׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               F!         X!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         	             2         3  G	         H	  \
         ]
  q         +  F         G  b         c  ~                                                                    &             K  E          p                                        !          D!  U"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                        ?         @  _         `                                                        	  
          
  ?         @  _         `                                                                      ?         @  _         `                                                                      ?         @  _         `                          !         !  "         "  $         $  )%         +%  L&         N&  o'         q'  (         (  )         )  *         *  +         +  -          -  A.         C.  d/         f/  0         0  1         1  2         2  3         3  5         5  66         86  Y7         [7  |8         ~8  9         9  :         :  ;         ;  =         
=  +>         ->  N?         P?  q@         s@  A         A  B         B  C         "D  CE         EE  fF         hF  G         G  H         I  K         K  $L         &L  GM         IM  jN         lN  O         O  P         P  Q         Q  R         R  T         T  <U         >U  _V         aV  W         W  X         X  Y         Y  Z         Z  \         \  1]         3]  T^         V^  w_         y_  `         `  a         a  b         b  d         d  &e         (e  If         Kf  lg          w  x          y  z          7z  3{          _{  [|          |  }                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     }            ,!          ^!  1          2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-g-     g--    -4-x    4-C-|    C-v-\    v--    --h    -9b-    9b-	n-	    	n--h
    -!׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               5!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        5  G          j  |                                  	            .  1          S  V	          x	  {
          
                                              -         .  H                      +         ,  @         A  U         V  j                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                     %  D         J  i         o                                              "	         (	  G
         M
  l         r                                               %         +  J         P  o         u                                            	  (         .  M         S  r         x                                                +!         1!  P"          )  *          *  +          +  -          4-  L.          p.  /         I  J         J  K         K  L         M  "N         (N  GO         MO  lP         rP  Q         Q  R         R  S         S   U         U  %V         +V  JW         PW  oX         uX  Y         Y  Z         Z  [         [  ]         	]  (^         .^  M_         S_  r`         x`  a         a  b         b  c         c  
e         e  0f         6f  Vg         \g  |h         h  i         i  j         j  k         k  m         m  :n         Dn  bo         lo  p         p  q         q  r         r  t         t  *u         4u  Rv         \v  zw         w  x         x  y         y  z         z  |         $|  B}         L}  j~         t~                      Ā             
           2         <  Z         d                        ҈         ܈             "         ,  J         T  r         |                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     q            !          H!  1          1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     C
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     --    -:-x    :-oI-|    oI-|-\    |--    --h    -ib-    ib-x-	    x--h
    -!׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               B!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         	             2         3  G	         H	  \
         ]
  q         +  E         F  `         a  {         |                                                           !            F  ?          k  }                                  
   !          ?!  Q"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                        >         @  ^         `  ~                                                      	  
          
  >         @  ^         `  ~                                                                    >         @  ^         `  ~                                                                    >         @  ^         `  ~                        !         !  "         "  $         $  *%         +%  M&         N&  p'         q'  (         (  )         )  *         *  +         +  -          -  B.         C.  e/         f/  0         0  1         1  2         2  3         3  5         5  76         86  Z7         [7  }8         ~8  9         9  :         :  ;         ;  	=         
=  ,>         ->  O?         P?  r@         s@  A         A  B         B  C         "D  DE         EE  gF         hF  G         G  H         I  K         K  %L         &L  HM         IM  kN         lN  O         O  P         P  Q         Q  R         R  T         T  =U         >U  `V         aV  W         W  X         X  Y         Y  Z         Z  \         \  2]         3]  U^         V^  x_         y_  `         `  a         a  b         b  d         d  'e         (e  Jf         Kf  mg          }  ~          ~                        '  #          O  K                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     |            +!          ^!  1          2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-s-     s--    -:-x    :-I-|    I-|-\    |--    --h    -?b-    ?b-x-	    x--h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               7!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         '  ;         <  P         Q  e	         f	  z
         {
           ?  ]         ^  |         }                                                                      7  &          V  E          u                                    &   8!          a!  s"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                        <         @  \         `  |                                                      	  
          
  <         @  \         `  |                                                                    <         @  \         `  |                                                                    <         @  \         `  |                        !         !  "         "  $         $  $%         +%  G&         N&  j'         q'  (         (  )         )  *         *  +         +  -          -  <.         C.  _/         f/  0         0  1         1  2         2  3         3  5         5  16         86  T7         [7  w8         ~8  9         9  :         :  ;         ;  =         
=  &>         ->  I?         P?  l@         s@  A         A  B         B  C         "D  >E         EE  aF         hF  G         G  H         I  J         K  L         &L  BM         IM  eN         lN  O         O  P         P  Q         Q  R         R  T         T  7U         >U  ZV         aV  }W         W  X         X  Y         Y  Z         Z  	\         \  ,]         3]  O^         V^  r_         y_  `         `  a         a  b         b  c         d  !e         (e  Df         Kf  gg          w  {x          x  y          y  z          z  {          $|  }                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     p            !          J!  1          1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 {
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-5-x    5-C-|    C-v-\    v-
-    
--h    -{b-    {b-x-	    x--h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               H!         X!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        ;  W          v                                    '  *          F  I          e  h	          	  
          
                                   <         =  e         f           *  >         ?  S         T  h         i  }         ~                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                             "  1          P  _         e$  %         %  &         &  '         '  (         (  *         
*  )+         ++  J,         L,  k-         m-  .         .  /         /  0         0  1         1  3         3  14         34  R5         T5  s6         u6  7         7  8         8  9         9  :         :  <         <  9=         ;=  Z>         \>  {?         }?  @         @  A         A  B         B  C         D   E         "E  AF         F  G         G  I         
I  )J         +J  JK         LK  kL         mL  M         M  N         N  O         O  P         P  R         R  1S         3S  RT         TT  sU         uU  V         V  W         W  X         X  Y         Y  [         [  9\         ;\  Z]         \]  {^         }^  _         _  `         `  a         a  b         c   d         "d  Ae         Ce  bf         df  g         g  h         h   j         "j  Dk         Fk  hl         jl  m         m  n         n  o         o  p         p  r         r  @s         Bs  dt         ft  u         u  v         v  w         w  x         x  z         z  <{         >{  `|         b|  }         }  ~         ~                                   8         :  \         ^                        ȇ         ʇ                        4                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               D!         T!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-m-     m--    -;-x    ;-I-|    I-|-\    |-O-    O--h    -Kb-    Kb-x-	    x--h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               5!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         '  ;         <  P         Q  e	         f	  z
         {
           D  c         d                                                                     "            A  /          `  N                                               0   C!          k!  ~"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                        =         @  ]         `  }                                                      	  
          
  =         @  ]         `  }                                                                    =         @  ]         `  }                                                                    =         @  ]         `  }                        !         !  "         "  $         $  %%         +%  H&         N&  k'         q'  (         (  )         )  *         *  +         +  -          -  =.         C.  `/         f/  0         0  1         1  2         2  3         3  5         5  26         86  U7         [7  x8         ~8  9         9  :         :  ;         ;  =         
=  '>         ->  J?         P?  m@         s@  A         A  B         B  C         "D  ?E         EE  bF         hF  G         G  H         I  J         K   L         &L  CM         IM  fN         lN  O         O  P         P  Q         Q  R         R  T         T  8U         >U  [V         aV  ~W         W  X         X  Y         Y  Z         Z  
\         \  -]         3]  P^         V^  s_         y_  `         `  a         a  b         b  c         d  "e         (e  Ef         Kf  hg          L}  C~          t~  k                      Ā                                 >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     q            !          N!  1          1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 {
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-5-x    5-C-|    C-w-\    w-
-    
--h    -ub-    ub-x-	    x--h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               D!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        ;  X          v                          	          '  )          F  H          e  g	          	  
          
                                   ?         @  i         j           /  C         D  X         Y  m         n                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         ~                                     !  "          6"  E#         e$  %         %  &         &  '         '  (         (  	*         
*  *+         ++  K,         L,  l-         m-  .         .  /         /  0         0  1         1  3         3  24         34  S5         T5  t6         u6  7         7  8         8  9         9  :         :  <         <  :=         ;=  [>         \>  |?         }?  @         @  A         A  B         B   D         D  !E         "E  BF         F  G         G  	I         
I  *J         +J  KK         LK  lL         mL  M         M  N         N  O         O  P         P  R         R  2S         3S  ST         TT  tU         uU  V         V  W         W  X         X  Y         Y  [         [  :\         ;\  []         \]  |^         }^  _         _  `         `  a         a   c         c  !d         "d  Be         Ce  cf         df  g         g  h         h  !j         "j  Ek         Fk  il         jl  m         m  n         n  o         o  p         p  r         r  As         Bs  et         ft  u         u  v         v  w         w  x         x  z         z  ={         >{  a|         b|  }         }  ~         ~                                   9         :  ]         ^                        ɇ         ʇ                        5                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               @!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-y-     y--    -4-x    4-C-|    C-v-\    v-
-    
--h    -Qb-    Qb-x-	    x--h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               9!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        5  F          j  {                                  	            .  2          S  W	          x	  |
          
                                              1         2  M                      0         1  E         F  Z         [  o                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         0  (          L  D          h  `            |                     e$  %         %  &         &  '         '  (         (  *         
*  $+         ++  E,         L,  f-         m-  .         .  /         /  0         0  1         1  3         3  ,4         34  M5         T5  n6         u6  7         7  8         8  9         9  :         :  <         <  4=         ;=  U>         \>  v?         }?  @         @  A         A  B         B  C         D  E         "E  <F         F  G         G  I         
I  $J         +J  EK         LK  fL         mL  M         M  N         N  O         O  P         P  R         R  ,S         3S  MT         TT  nU         uU  V         V  W         W  X         X  Y         Y  [         [  4\         ;\  U]         \]  v^         }^  _         _  `         `  a         a  b         c  d         "d  <e         Ce  ]f         df  ~g         g  h         h  j         "j  =k         Fk  al         jl  m         m  n         n  o         o  p         p  r         r  9s         Bs  ]t         ft  u         u  v         v  w         w  x         x  z         z  5{         >{  Y|         b|  }}         }  ~         ~                                   1         :  U         ^  y                               ʇ             	           -                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               -!         <!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 {
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-:-x    :-{I-|    {I-|-\    |-+-    +--h    -cb-    cb-!-	    !--h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               @!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         	             2         3  G	         H	  \
         ]
  q         +  F         G  b         c  ~                                                                    &             K  E          p                                        !          D!  U"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       7         8  S         T  o         p                                              	         	             3         4  O         P  k         l                                                                    /         0  K         L  g         h                                                                    +          ,   G!         !  "         "  #         #  %         %  :&         <&  Y'         ['  x(         z(  )         )  *         *  +         +  ,         ,  .         .  2/         4/  Q0         S0  p1         r1  2         2  3         3  4         4  5         5  7         7  *8         ,8  I9         K9  h:         j:  ;         ;  <         <  =         =  >         >  @         @  "A         $A  AB         CB  `C         C  E         E  4F         6F  SG         UG  rH         tH  I          $J  @K          dK  L          L  M          M   O          $O  @P         (  E         G  d         f                                 Õ                                    =         ?  \         ^  {         }                        ؟         ڟ                        5         7  T         V  s         u                        Ш         Ҩ                        -                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               3!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   S      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-s-     s--    -4-x    4-C-|    C-v-\    v-
-    
--h    -Wb-    Wb-x-	    x--h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               7!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        5  G          j  |                                  	            .  1          S  V	          x	  {
          
                                              -         .  H                      +         ,  @         A  U         V  j                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                                          ,   $!         e$  %         %  &         &  '         '  (         (  *         
*  %+         ++  F,         L,  g-         m-  .         .  /         /  0         0  1         1  3         3  -4         34  N5         T5  o6         u6  7         7  8         8  9         9  :         :  <         <  5=         ;=  V>         \>  w?         }?  @         @  A         A  B         B  C         D  E         "E  =F         F  G         G  I         
I  %J         +J  FK         LK  gL         mL  M         M  N         N  O         O  P         P  R         R  -S         3S  NT         TT  oU         uU  V         V  W         W  X         X  Y         Y  [         [  5\         ;\  V]         \]  w^         }^  _         _  `         `  a         a  b         c  d         "d  =e         Ce  ^f         df  g         g  h         h  j         "j  <k         Fk  `l         jl  m         m  n         n  o         o  p         p  r         r  8s         Bs  \t         ft  u         u  v         v  w         w  x         x  z         z  4{         >{  X|         b|  |}         }  ~         ~                                   0         :  T         ^  x                               ʇ                        ,                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               1!         @!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 {
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-:-x    :-uI-|    uI-|-\    |-+-    +--h    -cb-    cb-!-	    !--h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               @!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         	             2         3  G	         H	  \
         ]
  q         +  E         F  `         a  {         |                                                           !            F  ?          k  }                                  
   !          ?!  Q"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       6         8  R         T  n         p                                              	         	             2         4  N         P  j         l                                                                    .         0  J         L  f         h                                                                    *          ,   F!         !  "         "  #         #  %         %  ;&         <&  Z'         ['  y(         z(  )         )  *         *  +         +  ,         ,  .         .  3/         4/  R0         S0  q1         r1  2         2  3         3  4         4  5         5  7         7  +8         ,8  J9         K9  i:         j:  ;         ;  <         <  =         =  >         >  @         @  #A         $A  BB         CB  aC         C  E         E  5F         6F  TG         UG  sH         tH  I          dP  Q          Q  R          R   T          $T  @U          dU  V         (  F         G  e         f                                 Õ                                     >         ?  ]         ^  |         }                        ٟ         ڟ                        6         7  U         V  t         u                        Ѩ         Ҩ                        .                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               3!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   T      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-I-     I--    -:-x    :-I-|    I-|-\    |-+-    +--h    -cb-    cb-3-	    3--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                                !         0!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         '  ;         <  P         Q  e	         f	  z
         {
           ?  ]         ^  |         }                                                                      7  &          V  E          u                                    &   8!          a!  s"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       4         8  P         T  l         p                                              	         	             0         4  L         P  h         l                                                                    ,         0  H         L  d         h                                                                    (          ,   D!         !  "         "  #         #  %         %  5&         <&  T'         ['  s(         z(  )         )  *         *  +         +  ,         ,  .         .  -/         4/  L0         S0  k1         r1  2         2  3         3  4         4  5         5  7         7  %8         ,8  D9         K9  c:         j:  ;         ;  <         <  =         =  >         >  ?         @  A         $A  <B         CB  [C         C  E         E  /F         6F  NG         UG  mH         tH  I          I  J          J  K          L  -M          SM  mN          N  O         (  @         G  _         f  ~                               Õ  ۖ                                  8         ?  W         ^  v         }                        ӟ         ڟ                        0         7  O         V  n         u                        ˨         Ҩ             	           (                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               0!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   `      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-C<-x    C<-J-|    J-+~-\    +~--    --h    -Kb-    Kb-!-	    !--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               H!         V!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        7         8  S         T  o         p                                      .          R  c                                              2  N          m                                    =          >   f!         g!  "         "  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       :         <  X         Z  v         x                                              
         
  *         ,  H         J  f         h                                                                    8         :  V         X  t         v                                              
           (         *  F         H  d          f   !         #  $         $  &         &  :'         <'  Z(         \(  z)         |)  *         *  +         +  ,         ,  -         -  /         /  :0         <0  Z1         \1  z2         |2  3         3  4         4  5         5  6         6  8         8  :9         <9  Z:         \:  z;         |;  <         <  =         =  >         >  ?         ?  A         A  :B         <B  ZC         \C  zD         |D  E         F  G         G  H         H  J         J  ?K         @K  eL         fL  M         M  N         N  O         O  P         P  #R         $R  ES         GS  hT         jT  U         U  V         V  W         W  X         X  Z         Z  :[         <[  ]\         _\  ]         ]  ^         ^  _         _  `         `  b         b  %c         'c  Fd         Hd  ge         ie  f         f  g         g  h          Tv  sw          w  x          x  y          z  <{          `{  |                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ,!         6!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   S      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-C-     C--    -;-x    ;-I-|    I-|-\    |-+-    +--h    -]b-    ]b-3-	    3--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               !         .!  1         1  }=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         '  ;         <  P         Q  e	         f	  z
         {
           D  c         d                                                                     "            A  /          `  N                                               0   C!          k!  ~"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       5         8  Q         T  m         p                                              	         	             1         4  M         P  i         l                                                                    -         0  I         L  e         h                                                                    )          ,   E!         !  "         "  #         #  %         %  6&         <&  U'         ['  t(         z(  )         )  *         *  +         +  ,         ,  .         .  ./         4/  M0         S0  l1         r1  2         2  3         3  4         4  5         5  7         7  &8         ,8  E9         K9  d:         j:  ;         ;  <         <  =         =  >         >  ?         @  A         $A  =B         CB  \C         C  E         E  0F         6F  OG         UG  nH         tH  I          O  P          Q  -R          SR  mS          S  T          T  U         (  A         G  `         f                                 Õ  ܖ                                  9         ?  X         ^  w         }                        ԟ         ڟ                        1         7  P         V  o         u                        ̨         Ҩ             
           )                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               .!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   a      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-C<-x    C<-J-|    J-+~-\    +~-=-    =--h    -Qb-    Qb-'-	    '--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               D!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        5         6  P         Q  k         l                                      *          M  _                                              -  J          h                                    :          ;   d!         e!  "         "  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       ;         <  Y         Z  w         x                                              
         
  +         ,  I         J  g         h                                                                    9         :  W         X  u         v                                                         )         *  G         H  e          f   !         #  $         $  &         &  ;'         <'  [(         \(  {)         |)  *         *  +         +  ,         ,  -         -  /         /  ;0         <0  [1         \1  {2         |2  3         3  4         4  5         5  6         6  8         8  ;9         <9  [:         \:  {;         |;  <         <  =         =  >         >  ?         ?  A         A  ;B         <B  [C         \C  {D         |D  E         F  G         G  H         H  J         J  >K         @K  dL         fL  M         M  N         N  O         O  P         P  "R         $R  FS         GS  iT         jT  U         U  V         V  W         W  X         X  Z         Z  ;[         <[  ^\         _\  ]         ]  ^         ^  _         _  `         `  b         b  &c         'c  Gd         Hd  he         ie  f         f  g         g  h          |  }          }            )  H          l              ΂                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               .!         :!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   T      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-<-x    <-J-|    J-~-\    ~--    --h    -ob-    ob-3-	    3--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               A!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        =         >  \         ]  {         |                                      ,          U  g                                               5  F          j  {                                  2          3   N!         O!  j"         k"  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       5         <  S         Z  q         x                                              
         
  %         ,  C         J  a         h                                                                    3         :  Q         X  o         v                                                         #         *  A         H  _          f   }!         #  $         $  &         &  3'         <'  S(         \(  s)         |)  *         *  +         +  ,         ,  -         -  /         /  30         <0  S1         \1  s2         |2  3         3  4         4  5         5  6         6  8         8  39         <9  S:         \:  s;         |;  <         <  =         =  >         >  ?         ?  A         A  3B         <B  SC         \C  sD         |D  E         F  G         G  H         H  J         J  <K         @K  bL         fL  M         M  N         N  O         O  P         P   R         $R  @S         GS  cT         jT  U         U  V         V  W         W  X         X  Z         Z  5[         <[  X\         _\  {]         ]  ^         ^  _         _  `         `  a         b   c         'c  Ad         Hd  be         ie  f         f  g         g  h          t  u          u  v          w  x          <x  Fy          ly  vz                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               5!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   `      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-:-x    :-I-|    I-|-\    |-%-    %--P    -b-    b-c-	    c-m-D
    m-3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               E!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         	             2         3  G	         H	  \
         ]
  q         +  F         G  b         c  ~                                                                    &             K  E          p                                        !          D!  U"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p                        ?         @  _         `                                                        	  
          
  ?         @  _         `                                                                      ?         @  _         `                                                                      ?         @  _         `                          !         !  "         "  #          $  %          %  >&         @&  ^'         `'  ~(         (  )         )  *         *  +         +  ,          -  .          .  >/         @/  ^0         `0  ~1         1  2         2  3         3  4         4  5          6  7          7  >8         @8  ^9         `9  ~:         :  ;         ;  <         <  =         =  >          ?  @          @  >A         G  
I         I  *J         +J  JK         KK  jL         kL  M         M  N         N  O         O  P         P  
R         R  *S         +S  JT         KT  jU         kU  V         V  W         W  X         X  Y         Y  
[         [  *\         +\  J]         K]  j^         k^  _         _  `         `  a         a  b         b  d         d  6e         7e  Xf         Yf  zg         {g  h          ԅ                        8  I          j  {                               >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               A!         T!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-<-x    <-J-|    J-~-\    ~--    --h    -ob-    ob-3-	    3--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               E!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         ?         @  _         `                                                 2          Z  m                                              :  L          o                                     5          6   P!         Q!  k"         l"  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       6         <  T         Z  r         x                                              
         
  &         ,  D         J  b         h                                                                    4         :  R         X  p         v                                                         $         *  B         H  `          f   ~!         #  $         $  &         &  2'         <'  R(         \(  r)         |)  *         *  +         +  ,         ,  -         -  /         /  20         <0  R1         \1  r2         |2  3         3  4         4  5         5  6         6  8         8  29         <9  R:         \:  r;         |;  <         <  =         =  >         >  ?         ?  A         A  2B         <B  RC         \C  rD         |D  E         F  G         G  H         H  J         J  =K         @K  cL         fL  M         M  N         N  O         O  P         P  !R         $R  AS         GS  dT         jT  U         U  V         V  W         W  X         X  Z         Z  6[         <[  Y\         _\  |]         ]  ^         ^  _         _  `         `   b         b  !c         'c  Bd         Hd  ce         ie  f         f  g         g  h          z  {          {  |          |  ~          ,~  6          \  f                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               5!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   a      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-:-x    :-{I-|    {I-|-\    |--    --P    -b-    b-c-	    c-m-D
    m-3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               B!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         	             2         3  G	         H	  \
         ]
  q         +  E         F  `         a  {         |                                                           !            F  ?          k  }                                  
   !          ?!  Q"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p                        >         @  ^         `  ~                                                      	  
          
  >         @  ^         `  ~                                                                    >         @  ^         `  ~                                                                    >         @  ^         `  ~                        !         !  "         "  #          $  %          %  ?&         @&  _'         `'  (         (  )         )  *         *  +         +  ,          -  .          .  ?/         @/  _0         `0  1         1  2         2  3         3  4         4  5          6  7          7  ?8         @8  _9         `9  :         :  ;         ;  <         <  =         =  >          ?  @          @  ?A         G  	I         I  )J         +J  IK         KK  iL         kL  M         M  N         N  O         O  P         P  	R         R  )S         +S  IT         KT  iU         kU  V         V  W         W  X         X  Y         Y  	[         [  )\         +\  I]         K]  i^         k^  _         _  `         `  a         a  b         b  d         d  5e         7e  Wf         Yf  yg         {g  h          ΋  ߌ                       2  C          d  u                               >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               A!         T!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -/-    /-;-    ;-I-    I-|-d    |--    --0    --    -b-H	    b-c-P
    c-y-
    y-3׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               B!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         '  ;         <  P         Q  e	         f	  z
         {
           ?  ]         ^  |         }                                                                      7  &          V  E          u                                    &   8!          a!  s"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q   w  operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       \                        <         @  \         `  |                                                      	  
          
  <         @  \         `  |                                                                    <         @  \         `  |                                                                    <         @  \         `  |                        !         !  "         "  #          $  %          %  9&         @&  Y'         `'  y(         (  )         )  *         *  +         +  ,          -  .          .  9/         @/  Y0         `0  y1         1  2         2  3         3  4         4  5          6  7           7  8          @8  :9          `9  Z:          :  z;          ;  <         <  =         =  >          ?  @          @  9A         G  I         I  'J         +J  GK         KK  gL         kL  M         M  N         N  O         O  P         P  R         R  'S         +S  GT         KT  gU         kU  V         V  W         W  X         X  Y         Y  [         [  '\         +\  G]         K]  g^         k^  _         _  `         `  a         a  b         b  d         d  3e         7e  Uf         Yf  wg         {g  h                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               J!         l!   2         "2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -)-    )-y<-    y<-K-    K-a~-d    a~--    --p    -b-    b--
    -)-
    )--t    -3׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               B!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        7         8  S         T  o         p                                      .          R  c                                              2  I          m                                    O          P   !         !  "         "  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       &  +          L  Q          r  w                     $  8%         :%  [&         \&  }'         ~'  (         (  )         )  *         *  ,         ,  '-         (-  I.         J.  k/         l/  0         0  1         1  2         2  3         3  5         5  76         86  Y7         Z7  {8         |8  9         9  :         :  ;         ;  =         =  %>         &>  G?         H?  i@         j@  A         A  B         B  C         C  D         D  F         F  5G         6G  VH         XH  xI         zI  J         J  K         K  L         L  "N         $N  UO         WO  P         P  Q         Q  R         R  !T         #T  TU         VU  V         V  W         W  X         X  'Z         (Z  ][         ^[  \         \  ]         ]  ^          _  5`         6`  ka         la  b         b  c         c  e         e  ?f         @f  sg         tg  h         h  i         i  k         k  Cl         Dl  wm         xm  n         n  o         o   q         q  "r         $r  Ds         Fs  ft         ht  u         u  v         v  w         w  x         x  z         z  2{         4{  T|         V|  v}         x}  ~         ~             ܀         ހ                        "  B         D  d         f                        ʈ         ̈                        0                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               N!         b!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    \
         ]
             D      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     G
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -i -`     i -ְ-     ְ-1ܰ-    1ܰ-'4-x    '4-?-\    ?-Mh-<    Mh--    -?-    ?--    -b- 	    b-c-(
    c-y-
    y-3׳-
                   =         b      J      7   U   7   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               E!         R!  R1                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 64 })  outputElements  5184    input0Elements  31104   input1Elements  24576   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     4                           #         b      J      7   M     operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       @                                  *         +  ?	                                 /         0  O          P  >          o  ]            |                                              B  U          }                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     4                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  1728    input0Elements  1728    invocations     1                           3         b      J      7   Q   w  operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       \                        =         @  ]         `  }                                                      	  
          
  =         @  ]         `  }                                                                    =         @  ]         `  }                                                                    =         @  ]         `  }                        !         !  "         "  #          $  %          %  :&         @&  Z'         `'  z(         (  )         )  *         *  +         +  ,          -  .          .  :/         @/  Z0         `0  z1         1  2         2  3         3  4         4  5          6  7          7  :8         @8  Z9         `9  z:         :  ;         ;  <          <  =          =  >           ?  ?           @  A         G  I         I  (J         +J  HK         KK  hL         kL  M         M  N         N  O         O  P         P  R         R  (S         +S  HT         KT  hU         kU  V         V  W         W  X         X  Y         Y  [         [  (\         +\  H]         K]  h^         k^  _         _  `         `  a         a  b         b  d         d  4e         7e  Vf         Yf  xg         {g  h                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     4                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               H!         h!  1         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -)-    )-s<-    s<-K-    K-[~-d    [~--    --p    -b-    b--
    -)-
    )--t    -3׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               @!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        5         6  P         Q  k         l                                      *          M  _                                    
          -  E          h                                    H          I   y!         z!  "         "  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                 
  	          0	  5
          V
  [         $  9%         :%  Z&         \&  |'         ~'  (         (  )         )  *         *  ,         ,  &-         (-  H.         J.  j/         l/  0         0  1         1  2         2  3         3  5         5  66         86  X7         Z7  z8         |8  9         9  :         :  ;         ;  =         =  $>         &>  F?         H?  h@         j@  A         A  B         B  C         C  D         D  F         F  4G         6G  WH         XH  yI         zI  J         J  K         K  L         L  #N         $N  VO         WO  P         P  Q         Q  R         R  "T         #T  UU         VU  V         V  W         W  X         X  &Z         (Z  \[         ^[  \         \  ]         ]  ^          _  4`         6`  ja         la  b         b  c         c  
e         e  >f         @f  rg         tg  h         h  i         i  k         k  Bl         Dl  vm         xm  n         n  o         o  q         q  #r         $r  Es         Fs  gt         ht  u         u  v         v  w         w  x         x  z         z  3{         4{  U|         V|  w}         x}  ~         ~             ݀         ހ              !         "  C         D  e         f                        ˈ         ̈                        1                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               L!         b!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    \
         ]
             D      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     H
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -S-    S-1<-x    1<-J-|    J-~-\    ~-4-    4--h    -b-    b-n-	    n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               W!         X!  2         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             !         "  C         D  e         f                                               )  0          d  k                                               D  U          y                       
  %         &  A          B   ]!         ^!  y"         z"  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            $         &  J         L  p         r                                            
  .	         0	  T
         V
  z         |                                              8         :  ^         `                                                         B         D  h         j                                               &          (   L!         N!  r"          $  %          %  &           '  (          8(  O)          p)  *         K  L         L  M          N  %O         &O  KP         LP  qQ         rQ  R         R  S         S  T         T  	V         
V  /W         0W  UX         VX  {Y         |Y  Z         Z  [         [  \         \  ^         ^  9_         :_  _`         ``  a         a  b         b  c         c  d         d  f         f  Cg         Dg  ih         jh  i         i  j         j  k         k  m         m  'n         (n  Mo           ?         @  e         f                        י         ؙ             #         $  I         J  o         p                                              -         .  S         T  y         z             Ũ         ƨ                        7         8  ]         ^                        ϰ         а                        A         B  g                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               X!         b!  	2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     9
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     -A-    A-5-    5-C-    C-v-l    v-5-    5--x    -b-	    b--
    --
    -s-|    s-ճ-    ճ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               M!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        ;  O          v                                     '            F  9          e  X	          	  w
          
                                   $         %  E         F  f                     !  5         6  J         K  _         `  t                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |        8"  Y#         [#  |$         ~$  %         %  &         &  '         '  )         )  0*         2*  T+         V+  x,         z,  -         -  .         .  /         /  1         
1  ,2         .2  P3         R3  t4         v4  5         5  6         6  7         7  9         9  (:         *:  L;         N;  p<         r<  =         =  >         >  ?         ?   A         A  $B         &B  HC         JC  lD         nD  E         E  F         F  G          S  U          +U  =V          aV  sW          W  X          X  Y         -l  Nm         Pm  qn         sn  o         o  p         p  q         q  s         s  %t         't  Iu         Ku  mv         ov  w         w  x         x  y         y  z         z  !|         #|  E}         G}  i~         k~                        Ձ         ׁ                        A         C  e         g                        щ         Ӊ           ړ                       "  D         F  h         j                        ԛ         ֛                        @         B  d         f                        У         ң                        <         >  `         b                        ̫         Ϋ                        8         :  \         ^                        ȳ         ʳ                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               L!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    \
         ]
             "      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    \
         ]
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     g
            0     -,-X     ,--     -S-    S-+<-x    +<-J-|    J-~-\    ~-4-    4--h    -b-    b-n-	    n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               W!         X!  2         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                       !  A         B  b         c                                               $  ,          _  g                                  
            ?  Q          t                                     :          ;   U!         V!  p"         q"  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            %         &  K         L  q         r                                   	         
  /	         0	  U
         V
  {         |                                              9         :  _         `                                                         C         D  i         j                                              '          (   M!         N!  s"          *  +          +  ,          -  /.          P.  g/          /  0         K  L         L  M          N  $O         &O  JP         LP  pQ         rQ  R         R  S         S  T         T  V         
V  .W         0W  TX         VX  zY         |Y  Z         Z  [         [  \         \  ^         ^  8_         :_  ^`         ``  a         a  b         b  c         c  d         d  f         f  Bg         Dg  hh         jh  i         i  j         j  k         k   m         m  &n         (n  Lo           >         @  d         f                        ֙         ؙ             "         $  H         J  n         p                                              ,         .  R         T  x         z             Ĩ         ƨ                        6         8  \         ^                        ΰ         а                        @         B  f                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               Z!         f!  2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     8
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     -A-    A-5-    5-C-    C-w-l    w-5-    5--x    -b-	    b--
    --
    -y-|    y-ճ-    ճ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               M!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        ;  P          v                                    '            F  8          e  W	          	  v
          
                                   '         (  I         J  k           *         +  ?         @  T         U  i         j  ~                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |        8"  Z#         [#  }$         ~$  %         %  &         &  '         '  )         )  1*         2*  U+         V+  y,         z,  -         -  .         .  /         /  	1         
1  -2         .2  Q3         R3  u4         v4  5         5  6         6  7         7  9         9  ):         *:  M;         N;  q<         r<  =         =  >         >  ?         ?  A         A  %B         &B  IC         JC  mD         nD  E         E  F         F  G          Z  [          9[  K\          o\  ]          ]  ^          ^  _         -l  Om         Pm  rn         sn  o         o  p         p  q         q  s         s  &t         't  Ju         Ku  nv         ov  w         w  x         x  y         y  z         z  "|         #|  F}         G}  j~         k~                        ց         ׁ                        B         C  f         g                        ҉         Ӊ           ړ             !         "  E         F  i         j                        ՛         ֛                        A         B  e         f                        ѣ         ң                        =         >  a         b                        ͫ         Ϋ                        9         :  ]         ^                        ɳ         ʳ                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               J!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    \
         ]
             #      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    \
         ]
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     h
            0     -,-X     ,--     -A-    A-5-x    5-C-|    C-v-\    v--    --l    -b-    b-'n-     'n--l    -9׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               O!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        5  F          j  {                                  	            .  *          S  O	          x	  t
          
                                              1         2  M                      0         1  E         F  Z         [  o                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7         operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1                    >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               M!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     L
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *              operation.2     sync              "   432          |            "         $  F         H  j         l                                              	          	  B
         D
  f         h                                                         >         @  b         d                                                         :         <  ^         `                                              !         !  6"          /  1          51  T2          x2  3          3  4          4  6         .I  PJ         RJ  tK         vK  L         L  M         M  N         N  P         P  (Q         *Q  LR         NR  pS         rS  T         T  U         U  V         V   X         X  $Y         &Y  HZ         JZ  l[         n[  \         \  ]         ]  ^         ^  _         _  #a         $a  Ib         Jb  oc         pc  d         d  e         e  f         f  h         h  -i         .i  Sj         Tj  yk         m  n         n  p         p  <q         =q  `r         ar  s         s  t         t  u         u  v         v  x         x  8y         9y  \z         ]z  {         {  |         |  }         }  ~         ~             4         5  X         Y  |         }             ą         Ņ                        0         1  T         U  x         y                                            7            :            2            b         reason  ExchangeBarrier value   6       invocations     1             0     -,-X     ,--     -A-    A-+<-x    +<-J-|    J-~-\    ~--    --h    -b-    b--n-	    -n--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               L!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        7         8  S         T  o         p                                      .          R  c                                              2  F          m                                    -          .   N!         O!  o"         p"  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            $         &  J         L  p         r                                            
  .	         0	  T
         V
  z         |                                              8         :  ^         `                                                         B         D  h         j                                               &          (   L!         N!  r"         t"  #         #  $         $  %         %  '         '  1(         2(  W)         X)  }*         ~*  +         +  ,         ,  -         -  /         /  ;0         <0  a1         b1  2          K  L          M  #N          JN  fO          O  P          P  Q         _j  k         k  l         l  m         m  n         n  p         p  Aq         Cq  gr         ir  s         s  t         t  u         u  v         w  %x         'x  Ky         My  qz         sz  {         {  |         |  }         }  	           /         1  U         W  {         }             Ǆ         Ʉ                        9         ;  _         a                        ь         ӌ                      	  .         /  T         U  z         {             ƕ         Ǖ                        8         9  ^         _                        Н         ѝ                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               M!         Z!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     O
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -A-    A-5-x    5-C-|    C-v-\    v--    --h    -b-    b-'n-	    'n--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               O!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        5  G          j  |                                  	            .  )          S  N	          x	  s
          
                                              -         .  H                      +         ,  @         A  U         V  j                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            #         $  G         H  k         l                                              	          	  C
         D
  g         h                                                         ?         @  c         d                                                         ;         <  _         `                                              !         !  7"          A6  `7          7  8          8  9          
:  );          M;  l<         .I  QJ         RJ  uK         vK  L         L  M         M  N         N  P         P  )Q         *Q  MR         NR  qS         rS  T         T  U         U  V         V  X         X  %Y         &Y  IZ         JZ  m[         n[  \         \  ]         ]  ^         ^  _         _  "a         $a  Hb         Jb  nc         pc  d         d  e         e  f         f  h         h  ,i         .i  Rj         Tj  xk         m  n         n  p         p  ;q         =q  _r         ar  s         s  t         t  u         u  v         v  x         x  7y         9y  [z         ]z  {         {  |         |  }         }  ~         ~             3         5  W         Y  {         }             Å         Ņ                        /         1  S         U  w         y                                                      >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               M!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     K
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -{ -X     { -ְ-     ְ-Cܰ-    Cܰ-5-p    5-@-T    @-1i-4    1i--    --8    -b-    b--n-	    -n--8
    -9׳-
                   =         b      J      7   U   7   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               N!         T!  \1                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 64 })  outputElements  5184    input0Elements  31104   input1Elements  24576   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     4                           #         b      J      7   M     operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       @                        5         6  P         Q  k          X	  j
          
                          	          ,  A          g  |                                              /         0  Q         R  s                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     4                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  1728    input0Elements  1728    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       x            %         &  K         L  q         r                                   	         
  /	         0	  U
         V
  {         |                                              9         :  _         `                                                         C         D  i         j                                              '          (   M!         N!  s"         t"  #         #  $         $  %         %  
'         '  0(         2(  V)         X)  |*         ~*  +         +  ,         ,  -         -  /         /  :0         <0  `1         b1  2          R  /S          VS  rT          T  U          U  V         _j  k         k  l         l  m         m  n         n  p         p  Bq         Cq  hr         ir  s         s  t         t  u         u   w         w  &x         'x  Ly         My  rz         sz  {         {  |         |  }         }  
           0         1  V         W  |         }             Ȅ         Ʉ                        :         ;  `         a                        Ҍ         ӌ                      	  -         /  S         U  y         {             ŕ         Ǖ                        7         9  ]         _                        ϝ         ѝ                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               M!         Z!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     N
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -A-    A-#;-x    #;-I-|    I-}-\    }-+-    +--h    -b-    b-x-	    x--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               M!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         '  ;         <  P         Q  e	         f	  z
         {
           +  Q         R  x         y                                                          ,  +          K  J          j  i                                               :   T!          u!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                      "  B         D  d         f                                              	         	  0
         2
  R         T  t         v                                                          @         B  b         d                                                         .         0  P         R  r         t                                     !         "  1#         3#  S$          N;  Q<          r<  u=          =  >          >  ?          ?  @         sG  H         H  I         I  J         J  K         K  M         M  =N         ?N  _O         aO  P         P  Q         Q  R         R  S         S  	U         U  +V         -V  MW         OW  oX         qX  Y         Y  Z         Z  [         [  \         \  ^         ^  ;_         =_  ]`         _`  a         a  b         b  c         c  d         d  f         	f  )g         +g  Kh         Mh  mi         oi  j         j  k         k  l         $  D         F  f         h                        ̕         Ε                        2         4  T         V  v         x                        ܞ         ޞ                        "  B         D  d         f                        ʦ         ̦                        0         2  R         T  t                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               J!         V!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 {
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -;-    ;-5-x    5-C-|    C-v-\    v--    --h    -b-    b-x-	    x--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               L!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        ;  O          v                                     '            F  9          e  X	          	  w
          
                                   $         %  E         F  f                     !  5         6  J         K  _         `  t                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       *  -          T  W          ~                       "  
$         $  '%         ,%  G&         L&  g'         l'  (         (  )         )  *         *  +         +  -         -  '.         ,.  G/         L/  g0         l0  1         1  2         2  3         3  4         4  6         6  '7         ,7  G8         L8  g9         l9  :         :  ;         ;  <         <  =         =  ?         ?  '@         ,@  GA         LA  gB         lB  C         C  D         D  E         AF  \G         aG  |H         H  I         I  J         J  K         K  L         M  N         !N  <O         AO  \P         aP  |Q         Q  R         R  S         S  T         T  U         V  W         !W  <X         AX  \Y         aY  |Z         Z  [         [  \         \  ]         ]  ^         _  `         !`  <a         Aa  \b         ab  |c         c  d         d  e         e  f         f  g         h  i         i  k         k  0l         2l  Pm         Rm  pn         rn  o         o  p         p  q         q  r         r  t         t  0u         2u  Pv         Rv  pw         rw  x         x  y         y  z         z  {         {  }         }  0~         2~  P         R  p         r                        Ѓ         ҃                        0         2  P         R  p                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               L!         X!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-);-x    );-I-|    I-}-\    }-+-    +--h    -b-    b-x-	    x--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               I!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         '  ;         <  P         Q  e	         f	  z
         {
           +  R         S  z         {                                                          1  /          P  N          o  m                                               ?   Z!          z!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            !         "  C         D  e         f                                              	         	  1
         2
  S         T  u         v                                                          A         B  c         d                                                         /         0  Q         R  s         t                                     !         "  2#         3#  T$          A  B          &B  )C          JC  MD          nD  qE          E  F         sG  H         H  I         I  J         J  K         K  M         M  >N         ?N  `O         aO  P         P  Q         Q  R         R  S         S  
U         U  ,V         -V  NW         OW  pX         qX  Y         Y  Z         Z  [         [  \         \  ^         ^  <_         =_  ^`         _`  a         a  b         b  c         c  d         d  f         	f  *g         +g  Lh         Mh  ni         oi  j         j  k         k  l         $  E         F  g         h                        ͕         Ε                        3         4  U         V  w         x                        ݞ         ޞ              !         "  C         D  e         f                        ˦         ̦                        1         2  S         T  u                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               L!         Z!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 {
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -A-    A-5-x    5-C-|    C-v-\    v--    --h    -b-    b-x-	    x--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               N!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        ;  P          v                                    '            F  8          e  W	          	  v
          
                                   '         (  I         J  k           *         +  ?         @  T         U  i         j  ~                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                 &  )	          P	  S
          z
  }         "  $         $  (%         ,%  H&         L&  h'         l'  (         (  )         )  *         *  +         +  -         -  (.         ,.  H/         L/  h0         l0  1         1  2         2  3         3  4         4  6         6  (7         ,7  H8         L8  h9         l9  :         :  ;         ;  <         <  =         =  ?         ?  (@         ,@  HA         LA  hB         lB  C         C  D         D  E         AF  ]G         aG  }H         H  I         I  J         J  K         K  L         M  N         !N  =O         AO  ]P         aP  }Q         Q  R         R  S         S  T         T  U         V  W         !W  =X         AX  ]Y         aY  }Z         Z  [         [  \         \  ]         ]  ^         _  `         !`  =a         Aa  ]b         ab  }c         c  d         d  e         e  f         f  g         h  i         i  k         k  1l         2l  Qm         Rm  qn         rn  o         o  p         p  q         q  r         r  t         t  1u         2u  Qv         Rv  qw         rw  x         x  y         y  z         z  {         {  }         }  1~         2~  Q         R  q         r                        у         ҃                        1         2  Q         R  q                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               J!         T!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -;-    ;-4-x    4-C-|    C-v-\    v--    --h    -{b-    {b-x-	    x--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               M!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        5  F          j  {                                  	            .  *          S  O	          x	  t
          
                                              1         2  M                      0         1  E         F  Z         [  o                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                    <  0          Z  N          x  l         "  $         $  *%         ,%  J&         L&  j'         l'  (         (  )         )  *         *  +         +  
-         -  *.         ,.  J/         L/  j0         l0  1         1  2         2  3         3  4         4  
6         6  *7         ,7  J8         L8  j9         l9  :         :  ;         ;  <         <  =         =  
?         ?  *@         ,@  JA         LA  jB         lB  C         C  D         D  E         AF  _G         aG  H         H  I         I  J         J  K         K  L         M  N         !N  ?O         AO  _P         aP  Q         Q  R         R  S         S  T         T  U         V  W         !W  ?X         AX  _Y         aY  Z         Z  [         [  \         \  ]         ]  ^         _  `         !`  ?a         Aa  _b         ab  c         c  d         d  e         e  f         f  g         h  i         i  k         k  1l         2l  Qm         Rm  qn         rn  o         o  p         p  q         q  r         r  t         t  1u         2u  Qv         Rv  qw         rw  x         x  y         y  z         z  {         {  }         }  1~         2~  Q         R  q         r                        у         ҃                        1         2  Q         R  q                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               A!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 {
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -A-    A-:-x    :-I-|    I-|-\    |--    --h    -Kb-    Kb-'-	    '--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               N!         \!  1         
2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         	             2         3  G	         H	  \
         ]
  q         +  F         G  b         c  ~                                                                    &  (          K  M          p                                        !          D!  U"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       5         <  S         Z  q         x                                              
         
  %         ,  C         J  a         h                                                                    3         :  Q         X  o         v                                                         #         *  A         H  _          f   }!         p"  #         #  $         $  %         %  &         &  (         (  )         $)  =*         B*  [+         `+  y,         ~,  -         -  .         .  /         /  0         0  2         2  -3         23  K4         P4  i5         n5  6         6  7         7  8         8  9         9  :         ;  <         "<  ;=         @=  Y>         ^>  w?         |?  @         @  A         A  B         B  C         C  E         E  +F         0F  IG         :H  SI         XI  qJ         vJ  K         K  L         L  M         M  N         N  P         P  %Q         *Q  CR         HR  aS         fS  T         T  U         U  V         V  W         W  X         X  Z         Z  3[         8[  Q\         V\  o]         t]  ^         ^  _         _  `         `  a         a  c         
c  #d         (d  Ae         Fe  _f          v  w          w  x          x  z          *z  8{          \{  j|                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               %!         8!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   S      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-4-x    4-C-|    C-v-\    v--    --h    -{b-    {b-x-	    x--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               I!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        5  G          j  |                                  	            .  )          S  N	          x	  s
          
                                              -         .  H                      +         ,  @         A  U         V  j                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                               	          
           "  
$         $  +%         ,%  K&         L&  k'         l'  (         (  )         )  *         *  +         +  -         -  +.         ,.  K/         L/  k0         l0  1         1  2         2  3         3  4         4  6         6  +7         ,7  K8         L8  k9         l9  :         :  ;         ;  <         <  =         =  ?         ?  +@         ,@  KA         LA  kB         lB  C         C  D         D  E         AF  `G         aG  H         H  I         I  J         J  K         K   M         M   N         !N  @O         AO  `P         aP  Q         Q  R         R  S         S  T         T   V         V   W         !W  @X         AX  `Y         aY  Z         Z  [         [  \         \  ]         ]   _         _   `         !`  @a         Aa  `b         ab  c         c  d         d  e         e  f         f   h         h  i         i  k         k  0l         2l  Pm         Rm  pn         rn  o         o  p         p  q         q  r         r  t         t  0u         2u  Pv         Rv  pw         rw  x         x  y         y  z         z  {         {  }         }  0~         2~  P         R  p         r                        Ѓ         ҃                        0         2  P         R  p                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               A!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 {
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -;-    ;-:-x    :-{I-|    {I-|-\    |--    --h    -Qb-    Qb-'-	    '--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               L!         X!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         	             2         3  G	         H	  \
         ]
  q         +  E         F  `         a  {         |                                                           !  "          F  G          k  }                                  
   !          ?!  Q"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       4         <  R         Z  p         x                                              
         
  $         ,  B         J  `         h  ~                                                                  2         :  P         X  n         v                                                         "         *  @         H  ^          f   |!         p"  #         #  $         $  %         %  &         &  (         (   )         $)  >*         B*  \+         `+  z,         ~,  -         -  .         .  /         /  0         0  2         2  .3         23  L4         P4  j5         n5  6         6  7         7  8         8  9         9   ;         ;  <         "<  <=         @=  Z>         ^>  x?         |?  @         @  A         A  B         B  C         C  E         E  ,F         0F  JG         :H  TI         XI  rJ         vJ  K         K  L         L  M         M  N         N  P         P  &Q         *Q  DR         HR  bS         fS  T         T  U         U  V         V  W         W  X         X  Z         Z  4[         8[  R\         V\  p]         t]  ^         ^  _         _  `         `  a         a  c         
c  $d         (d  Be         Fe  `f          |  }          }  ~          ~             $  2          V  d                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               )!         <!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   T      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-y-     y--    -;-x    ;-I-|    I-}-\    }-1-    1--h    -ub-    ub-9-	    9--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ;!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         '  ;         <  P         Q  e	         f	  z
         {
           +  Q         R  x         y                                                          ,  +          K  J          j  i                                               :   T!          u!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       :         <  X         Z  v         x                                              
         
  *         ,  H         J  f         h                                                                    8         :  V         X  t         v                                              
           (         *  F         H  d          f   !         p"  #         #  $         $  %         %  &         &  (         (  ")         $)  @*         B*  ^+         `+  |,         ~,  -         -  .         .  /         /  0         0  2         2  03         23  N4         P4  l5         n5  6         6  7         7  8         8  9         9  ;         ;   <         "<  >=         @=  \>         ^>  z?         |?  @         @  A         A  B         B  C         C  E         E  .F         0F  LG         :H  VI         XI  tJ         vJ  K         K  L         L  M         M  N         N  
P         P  (Q         *Q  FR         HR  dS         fS  T         T  U         U  V         V  W         W  X         X  Z         Z  6[         8[  T\         V\  r]         t]  ^         ^  _         _  `         `  a         a  c         
c  &d         (d  De         Fe  bf          y  z          z  {          |  9}          Z}  |~          ~                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               :!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   h      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-<-x    <-J-|    J-~-\    ~--    --h    -9b-    9b-!-	    !--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               @!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        7         8  S         T  o         p                                      .          R  c                                              2  F          m                                    -          .   N!         O!  o"         p"  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       5         :  R         W  o         t                                               
         
           "  :         ?  W         \  t         y                                                       
  "         '  ?         D  \         a  y         ~                                              
           '         ,  D          I   a!         d#  ~$         $  %         %  &         &  '         '  )         )  '*         ,*  K+         P+  o,         t,  -         -  .         .  /         /  0         1  !2         &2  C3         H3  e4         j4  5         5  6         6  7         7  8         8  :         :  1;         6;  S<         X<  s=         x=  >         >  ?         ?  @         @  A         A  C         C  3D         8D  SE         XE  sF         xF  G          b  
d          .d  Me          qe  f          f  g          g  i         m  n         n  9o         ;o  Up         Wp  qq         sq  r         r  s         s  t         t  u         u  v         v  x         x  5y         7y  Qz         Sz  m{         o{  |         |  }         }  ~         ~                                   1         3  M         O  i         k                                   ى         ۉ                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                                !         .!  1         1  ~=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   S      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     --    -#;-x    #;-I-|    I-}-\    }--    --h    -ob-    ob-9-	    9--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               =!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         '  ;         <  P         Q  e	         f	  z
         {
           +  R         S  z         {                                                          1  /          P  N          o  m                                               ?   Z!          z!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       ;         <  Y         Z  w         x                                              
         
  +         ,  I         J  g         h                                                                    9         :  W         X  u         v                                                         )         *  G         H  e          f   !         p"  #         #  $         $  %         %  &         &  (         (  #)         $)  A*         B*  _+         `+  },         ~,  -         -  .         .  /         /  0         0  2         2  13         23  O4         P4  m5         n5  6         6  7         7  8         8  9         9  ;         ;  !<         "<  ?=         @=  ]>         ^>  {?         |?  @         @  A         A  B         B  C         C  E         E  /F         0F  MG         :H  WI         XI  uJ         vJ  K         K  L         L  M         M  N         N  P         P  )Q         *Q  GR         HR  eS         fS  T         T  U         U  V         V  W         W  X         X  Z         Z  7[         8[  U\         V\  s]         t]  ^         ^  _         _  `         `  a         a  	c         
c  'd         (d  Ee         Fe  cf                      #  E          f              ˄                               >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               6!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   i      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     --    -<-x    <-J-|    J-~-\    ~-a-    a--h    -?b-    ?b-'-	    '--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               >!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        5         6  P         Q  k         l                                      *          M  _                                              -  B          h  }                                	  *          +   L!         M!  n"         o"  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       6         :  S         W  p         t                                              
         
           "  ;         ?  X         \  u         y                                                       
  #         '  @         D  ]         a  z         ~                                                         (         ,  E          I   b!         d#  $         $  %         %  &         &  '         '  )         )  (*         ,*  L+         P+  p,         t,  -         -  .         .  /         /   1         1  "2         &2  D3         H3  f4         j4  5         5  6         6  7         7  8         8  :         :  2;         6;  T<         X<  t=         x=  >         >  ?         ?  @         @  A         A  C         C  4D         8D  TE         XE  tF         xF  G          :i  Yj          }j  k          k  l         m  n         n  :o         ;o  Vp         Wp  rq         sq  r         r  s         s  t         t  u         u  v         v  x         x  6y         7y  Rz         Sz  n{         o{  |         |  }         }  ~         ~                                   2         3  N         O  j         k                                   ډ         ۉ                                   э                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               $!         2!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   T      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-7<-x    7<-J-|    J-~-\    ~--    --h    -Wb-    Wb-3-	    3--h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               I!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             &         '  M         N  t         u                                     !          B  \          }                                    (  9          ]  n                                 #  >         ?  Z          [   v!         w!  "         "  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       8         :  U         W  r         t                                              
         
            "  =         ?  Z         \  w         y                                                       
  %         '  B         D  _         a  |         ~                                                         *         ,  G          I   d!         d#  $         $  %         %  &         &  '         '  )         )  **         ,*  N+         P+  r,         t,  -         -  .         .  /         /  1         1  $2         &2  F3         H3  h4         j4  5         5  6         6  7         7  8         8  :         :  4;         6;  V<         X<  v=         x=  >         >  ?         ?  @         @  A         A  C         C  6D         8D  VE         XE  vF         xF  G          `  a          a  b          
c  *d          Hd  he          e  f         m  n         n  :o         ;o  Vp         Wp  rq         sq  r         r  s         s  t         t  u         u  v         v  x         x  6y         7y  Rz         Sz  n{         o{  |         |  }         }  ~         ~                                   2         3  N         O  j         k                                   ډ         ۉ                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               1!         >!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   h      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-s-     s--    -:-x    :-{I-|    {I-|-\    |--    --P    -ib-    ib-c-	    c-m-D
    m-'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               7!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         	             2         3  G	         H	  \
         ]
  q         +  F         G  b         c  ~                                                                    &  (          K  M          p                                        !          D!  U"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p                     "  =         D  _         f                                              		         	  +
         2
  M         T  o         v                                                          ;         B  ]         d                                                         )         0  K         R  m         t                                     !         !  #         #  2$         7$  Q%         V%  p&         u&  '         '  (         (  )         )  +         +  6,         =,  ^-         e-  .         .  /         /  0         0  1         2  (3         /3  R4         Y4  |5         5  6         6  7         7  8         9  $:         +:  N;         U;  x<         <  =         =  >         >  ?         @  &A         -A  RB         YB  ~C         C  D         D  E         E  G         	G  .H         5H  OI          TI  aJ          J  K          n  o         o  9p         >p  Xq         ]q  wr         |r  s         s  t         t  u         u  v         v  x         x  1y         6y  Pz         Uz  o{         t{  |         |  }         }  ~         ~             
           )         .  H         M  g         l                        ć                       A  N          o  |                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               9!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-7<-x    7<-J-|    J-~-\    ~--    --h    -Qb-    Qb-3-	    3--h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               G!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             '         (  O         P  w         x                                     '          G  b                                    
          -  ?          b  t                                 (  B         C  ]          ^   x!         y!  "         "  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       9         :  V         W  s         t                                              
         
  !         "  >         ?  [         \  x         y                                              	         
  &         '  C         D  `         a  }         ~                                                         +         ,  H          I   e!         d#  ~$         $  %         %  &         &  '         '  )         )  +*         ,*  O+         P+  s,         t,  -         -  .         .  /         /  1         1  %2         &2  G3         H3  i4         j4  5         5  6         6  7         7  8         8  :         :  5;         6;  W<         X<  w=         x=  >         >  ?         ?  @         @  A         A  C         C  7D         8D  WE         XE  wF         xF  G          f  g          h  "i          @i  `j          ~j  k          k  l         m  n         n  9o         ;o  Up         Wp  qq         sq  r         r  s         s  t         t  u         u  v         v  x         x  5y         7y  Qz         Sz  m{         o{  |         |  }         }  ~         ~                                   1         3  M         O  i         k                                   ى         ۉ                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               /!         >!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   i      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-m-     m--    -:-x    :-uI-|    uI-|-\    |--    --P    -ib-    ib-c-	    c-m-D
    m-'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               6!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         	             2         3  G	         H	  \
         ]
  q         +  E         F  `         a  {         |                                                           !  "          F  G          k  }                                  
   !          ?!  Q"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p                     "  <         D  ^         f                                              	         	  *
         2
  L         T  n         v                                                          :         B  \         d  ~                                                       (         0  J         R  l         t                                     !         !  #         #  3$         7$  R%         V%  q&         u&  '         '  (         (  )         )  +         +  5,         =,  ]-         e-  .         .  /         /  0         0  1         2  '3         /3  Q4         Y4  {5         5  6         6  7         7  8         9  #:         +:  M;         U;  w<         <  =         =  >         >  ?         @  %A         -A  QB         YB  }C         C  D         D  E         E  G         	G  -H         5H  PI          n  o         o  :p         >p  Yq         ]q  xr         |r  s         s  t         t  u         u  v         v  x         x  2y         6y  Qz         Uz  p{         t{  |         |  }         }  ~         ~                        *         .  I         M  h         l                        Ň                      ː  ؑ                      '  4          U  b                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               9!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -)-    )-;-    ;-I-    I-|-d    |-/-    /--0    --    -b-H	    b-c-P
    c-y-
    y-'׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               F!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         '  ;         <  P         Q  e	         f	  z
         {
           +  Q         R  x         y                                                          ,  +          K  J          j  i                                               :   T!          u!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q   w  operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       \                      "  B         D  d         f                                              	         	  0
         2
  R         T  t         v                                                          @         B  b         d                                                         .         0  P         R  r         t                                      !          !  "          #  $          7$  5%          V%  T&         u&  '         '  (         (  )         )  +         +  ;,         =,  c-         e-  .         .  /         /  0         0  2         2  -3         /3  W4         Y4  5         5  6         6  7         7  8         9  ):         +:  S;         U;  }<         <  =         =  >         >  ?         @  +A         -A  WB         YB  C         C  D         D  E         E  G         	G  3H         5H  RI          n  o         o  <p         >p  [q         ]q  zr         |r  s         s  t         t  u         u  v         v  x         x  4y         6y  Sz         Uz  r{         t{  |         |  }         }  ~         ~                        ,         .  K         M  j         l                        Ǉ                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               N!         \!  1         
2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,-a-     a--    -<-    <-J-    J-}-l    }-	-    	--P    --    -b-h	    b-낳-p
    낳--
    -۰-    ۰-'׳-L                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               2!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        7         8  S         T  o         p                                      .          R  c                                              2  F          m                                    -          .   N!         O!  o"         p"  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        #  $         $  %         %  &         '  #(         *(  G)         N)  k*         r*  +         +  ,         ,  -         -  .         /  0         &0  C1         J1  g2         n2  3         3  4         4  5         5  6         6  8         "8  ?9         F9  c:         j:  ;         ;  <         <  =         =  >         >  @         @  ;A         BA  _B         fB  C         C  D         D  E         E  F         F  H         H  6I         ;I  WJ         \J  xK         }K  L         L  M         M  N         N  P         
P  ,Q         .Q  PR         RR  tS         vS  T         T  U         U  V         V  X         X  &Y         (Y  HZ         JZ  j[         l[  \         \  ]         ]  ^         ^  _         _  a         a  6b         8b  Xc         Zc  xd         zd  e         e  f         f  g         g  h         h  j         j  8k         :k  Xl         Zl  xm         zm  n         n  o         o  p         p  q         q  s          s  t          ?t  <u          `u  ]v          v  ~w          w  x         x  y         y   {         {  !|         &|  B}         G}  c~         h~                        Ɓ         ˁ                        )         .  J         O  k         p                                           b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               P!         Z!  1         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    \
         ]
             "                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     G
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -#-    #-;-    ;-I-    I-}-d    }-/-    /--0    --    -b-H	    b-c-P
    c-y-
    y-'׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               E!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         '  ;         <  P         Q  e	         f	  z
         {
           +  R         S  z         {                                                          1  /          P  N          o  m                                               ?   Z!          z!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q   w  operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       \            !         "  C         D  e         f                                              	         	  1
         2
  S         T  u         v                                                          A         B  c         d                                                         /         0  Q         R  s         t                                     !         !  #         #  6$         7$  U%         V%  t&         u&  '         '  (         (  )         )  +         +  <,         =,  d-         e-  .         .  /         /  0         0  2         2  .3         /3  X4         Y4  5         5  6         6  7         7   9         9  *:         +:  T;         U;  ~<         <  =         =  >         >   @         @  ,A         -A  XB         YB  C         C  D         D  E         E  G         	G  4H          5H  3I           n  n          o  p          >p  <q          ]q  [r         |r  s         s  t         t  u         u  v         v  x         x  5y         6y  Tz         Uz  s{         t{  |         |  }         }  ~         ~                        -         .  L         M  k         l                        ȇ                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               L!         X!  1         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,-a-     a--    -<-    <-J-    J-}-l    }-	-    	--P    --    -b-h	    b-낳-p
    낳--
    -۰-    ۰-'׳-L                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               2!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        5         6  P         Q  k         l                                      *          M  _                                              -  B          h  }                                	  *          +   L!         M!  n"         o"  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        #  $         $  %         %  &         '  "(         *(  F)         N)  j*         r*  +         +  ,         ,  -         -  .         /  0         &0  B1         J1  f2         n2  3         3  4         4  5         5  6         6  8         "8  >9         F9  b:         j:  ;         ;  <         <  =         =  >         >  @         @  :A         BA  ^B         fB  C         C  D         D  E         E  F         F  H         H  7I         ;I  XJ         \J  yK         }K  L         L  M         M  N         N  	P         
P  -Q         .Q  QR         RR  uS         vS  T         T  U         U  V         V  X         X  'Y         (Y  IZ         JZ  k[         l[  \         \  ]         ]  ^         ^  _         _  a         a  7b         8b  Yc         Zc  yd         zd  e         e  f         f  g         g  h         h  j         j  9k         :k  Yl         Zl  ym         zm  n         n  o         o  p         p  q         q  s         s  ;t         ?t  \u         `u  }v         v  w         w  x          x  y          y  z          {  |          &|  #}          G}  D~         h~                        ǁ         ˁ             	           *         .  K         O  l         p                                           b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               N!         Z!  1         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    \
         ]
             #                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     H
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,-y-     y--    -1<-    1<-J-    J-~-d    ~-	-    	--p    -b-    b-낳-
    낳--
    -ᰳ-t    ᰳ-!׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               A!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             &         '  M         N  t         u                                     !          B  \          }                                    (  9          ]  n                                 #  >         ?  Z          [   v!         w!  "         "  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                             "  +          L  U         #  $         $  %         %  '         '  ((         *(  L)         N)  p*         r*  +         +  ,         ,  -         -   /         /  $0         &0  H1         J1  l2         n2  3         3  4         4  5         5  6         6   8         "8  D9         F9  h:         j:  ;         ;  <         <  =         =  >         >  @         @  @A         BA  dB         fB  C         C  D         D  E         E  F         F  H         H  9I         ;I  ZJ         \J  {K         }K  L         L  M         M  N         N  	P         
P  -Q         .Q  QR         RR  uS         vS  T         T  U         U  V         V  X         X  'Y         (Y  IZ         JZ  k[         l[  \         \  ]         ]  ^         ^  _         _  a         a  7b         8b  Yc         Zc  yd         zd  e         e  f         f  g         g  h         h  j         j  9k         :k  Yl         Zl  ym         zm  n         n  o         o  p         p  q         q  s         s  =t         ?t  ^u         `u  v         v  w         w  x         x  y         y  {         {  $|         &|  E}         G}  f~         h~                        Ɂ         ˁ                        ,         .  M         O  n         p                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               a!         f!  2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    \
         ]
             $      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     P
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                 w        Sܯ--0     -y-     y-]-     ]-	B-P    	B-c-    c--    -!׳-p                             b      *      7   I      operation.0     sync             :            2            b         reason  ExchangeBarrier value   1       invocations     1                                     b      *      7   I      operation.0     sync             :            2            b         reason  ExchangeBarrier value   3       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   5       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   2       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,-y-     y--    -1<-    1<-J-    J-~-d    ~-	-    	--p    -b-    b-낳-
    낳--
    -ᰳ-t    ᰳ-!׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               A!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             '         (  O         P  w         x                                     '          G  b                                    
          -  ?          b  t                                 (  B         C  ]          ^   x!         y!  "         "  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         v                                                  '         #  $         $  %         %  '         '  )(         *(  M)         N)  q*         r*  +         +  ,         ,  -         -  /         /  %0         &0  I1         J1  m2         n2  3         3  4         4  5         5  6         6  !8         "8  E9         F9  i:         j:  ;         ;  <         <  =         =  >         >  @         @  AA         BA  eB         fB  C         C  D         D  E         E  F         F  H         H  :I         ;I  [J         \J  |K         }K  L         L  M         M  N         N  P         
P  ,Q         .Q  PR         RR  tS         vS  T         T  U         U  V         V  X         X  &Y         (Y  HZ         JZ  j[         l[  \         \  ]         ]  ^         ^  _         _  a         a  6b         8b  Xc         Zc  xd         zd  e         e  f         f  g         g  h         h  j         j  8k         :k  Xl         Zl  xm         zm  n         n  o         o  p         p  q         q  s         s  >t         ?t  _u         `u  v         v  w         w  x         x  y         y  {         {  %|         &|  F}         G}  g~         h~                        ʁ         ˁ                        -         .  N         O  o         p                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               a!         f!  2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    \
         ]
             $      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     Q
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                 w        Sܯ--0     -y-     y-]-     ]-	B-P    	B-c-    c--    -!׳-p                             b      *      7   I      operation.0     sync             :            2            b         reason  ExchangeBarrier value   1       invocations     1                                     b      *      7   I      operation.0     sync             :            2            b         reason  ExchangeBarrier value   3       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   5       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   2       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    <     -,-p     ,-7-     7--    -4-    4-C-    C-v-t    v-30-     30-7-X    7--    -ib-p	    ib-傳-x
    傳--
    -U-    U-ճ-\    ճ-׳-<                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     l            !          F!  1          1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        3  @          f  s                                              &  *          M  Q	          t	  x
          
                                              )         *  C                      4         5  K         L  b         c  y                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        T$  v%         x%  &         vL  M         M  N         N  O         O  Q         Q  )R         *R  MS         NS  qT         rT  U         U  V         V  W         W  Y         Y  %Z         &Z  I[         J[  m\         n\  ]         ]  ^         ^  _         _  `         `  !b         "b  Ec         Fc  id         jd  e         e  f         f  g         g  h         h  j         j  Ak         Bk  el         fl  m         m  n         n  o         o  p         p  r         r  =s         >s  at         bt  u         u  v         v  w         w  x         x  z         z  9{         :{  ]|         ^|  }         }  ~         ~                                   5         6  Y         Z  }         ~             Ň         Ƈ                        1         2  U         V  y         z                                            
  ,         .  P         R  t         v                                              (         *  L         N  p         r                        ܠ         ޠ              $         &  H         J  l          n  o                                  ڨ  ۩                     "  D         F  h         j                        ԰                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               G!         H!  1         1  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    ^
         _
             &                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    ^
         _
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,-U-     U--    -1<-    1<-J-    J-~-d    ~-[-    [--p    -b-    b--
    --
    -ɰ-t    ɰ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     y             !          N!  1          1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        3         4  M         N  g         h                                      &          L  Y                                              ,  H          i                                    7          8   b!         c!  "         "  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                                         "  &         $  %         %  &         &  (         (  ')         ()  M*         N*  s+         t+  ,         ,  -         -  .         .  0         0  11         21  W2         X2  }3         ~3  4         4  5         5  6         6  8         8  ;9         <9  a:         b:  ;         ;  <         <  =         =  >         >  @          @  EA         FA  kB         lB  C         C  D         D  E         E  G         G  (H         *H  NI         PI  tJ         vJ  K         K  L         L  M         M  O         O  2P         4P  XQ         ZQ  ~R         R  S         S  T         T  U         U  W         W  <X         >X  bY         dY  Z         Z  [         [  \         \  ]         ]   _         "_  F`         H`  la         na  b         b  c         c  d         d  f         
f  0g         2g  Xh         Zh  i         i  j         j  k         k  l         l   n         "n  Ho         Jo  np         pp  q         q  r         r  s         s  u         u  ,v         .v  Rw         Tw  xx         zx  y         y  z         z  {         {  }         }  6~         8~  \         ^                        ΂         Ђ                        @         B  f         h                        ؊         ڊ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               e!         f!  2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    ^
         _
             2      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     C
                   
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            <     -,-p     ,-C-     C--    -4-    4-C-    C-v-t    v-0-     0-7-X    7--    -ib-p	    ib-傳-x
    傳--
    -U-    U-ճ-\    ճ-׳-<                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     m            !          J!  1          1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        3  A          f  t                                              &  )          M  P	          t	  w
          
                                              %         &  >                      /         0  F         G  ]         ^  t                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        T$  w%         x%  &         vL  M         M  N         N  O         O  Q         Q  (R         *R  LS         NS  pT         rT  U         U  V         V  W         W   Y         Y  $Z         &Z  H[         J[  l\         n\  ]         ]  ^         ^  _         _  `         `   b         "b  Dc         Fc  hd         jd  e         e  f         f  g         g  h         h  j         j  @k         Bk  dl         fl  m         m  n         n  o         o  p         p  r         r  <s         >s  `t         bt  u         u  v         v  w         w  x         x  z         z  8{         :{  \|         ^|  }         }  ~         ~                                   4         6  X         Z  |         ~             ć         Ƈ                        0         2  T         V  x         z                                   	         
  -         .  Q         R  u         v                                              )         *  M         N  q         r                        ݠ         ޠ             %         &  I         J  m         n                        ٨         ڨ             !          "  #          F  G          j  k                                                      b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               G!         H!  1         1  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    ^
         _
             &                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    ^
         _
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,-O-     O--    -1<-    1<-J-    J-~-d    ~-[-    [--p    -b-    b--
    --
    -ϰ-t    ϰ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     x            !          N!  1          1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        1         2  J         K  c         d  |                                    "          G  U          z                                    '  D          d                                  	  4          5   `!         a!  "         "  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         I  M          p  t                                             $  %         %  &         &   (         (  &)         ()  L*         N*  r+         t+  ,         ,  -         -  .         .  
0         0  01         21  V2         X2  |3         ~3  4         4  5         5  6         6  8         8  :9         <9  `:         b:  ;         ;  <         <  =         =  >         >  @          @  DA         FA  jB         lB  C         C  D         D  E         E  G         G  )H         *H  OI         PI  uJ         vJ  K         K  L         L  M         M  O         O  3P         4P  YQ         ZQ  R         R  S         S  T         T  U         U  W         W  =X         >X  cY         dY  Z         Z  [         [  \         \  ]         ]  !_         "_  G`         H`  ma         na  b         b  c         c  d         d  	f         
f  1g         2g  Yh         Zh  i         i  j         j  k         k  l         l  !n         "n  Io         Jo  op         pp  q         q  r         r  s         s  u         u  -v         .v  Sw         Tw  yx         zx  y         y  z         z  {         {  }         }  7~         8~  ]         ^                        ς         Ђ                        A         B  g         h                        ي         ڊ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               e!         f!  2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    ^
         _
             4      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     D
                   
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,-C-     C--    -<-x    <-J-|    J-}-\    }-]3-    ]3--h    -ub-    ub-n-	    n--h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     l            !          H!  1          1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                       !  A         B  b         c                                                 0          [  m                                              ;  H          n  {                                  8          9   R!         S!  l"         m"  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                      '  G         N  n         u                                   
           1	         8	  X
         _
                                                       "  B         I  i         p                                              ,         3  S         Z  z                                                       =          D   d!         k!  "          %  &          '  :(          `(  })          )  *          *  ,         K  L         L  N         N  5O         >O  cP         lP  Q         Q  R         R  S         S  U         $U  IV         RV  wW         W  X         X  Y         Y  Z         [  '\         0\  S]         \]  ^         ^  _         _  `         `  b         b  /c         8c  Yd         bd  e         e  f         f  g         g  i         
i  +j         4j  Uk         ^k  l         l  m         m  n         ȑ                        0         4  T         X  x         |                      ę                        ,         0  P         T  t         x                                              (         ,  L         P  p         t                        ܪ                       $         (  H         L  l         p                        ز         ܲ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               J!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     @
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            8     -,-h     ,-y-     y--    -5-    5-C-    C-w-l    w---    ---x    -Qb-	    Qb--
    --
    -m-|    m-ճ-    ճ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               A!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        =  Y          z                                    1  0          N  M          k  j	          	  
          
                                   B         C  m         n           *  @         A  W         X  n         o                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |        V"  {#         }#  $          <  =          =  >          >  @          "@  6A          TA  hB         H  I         I  J         J  K         K  M         M  EN         GN  mO         oO  P         P  Q         Q  R         R  T         T  5U         7U  ]V         _V  W         W  X         X  Y         Y  Z         Z  %\         '\  M]         O]  u^         w^  _         _  `         `  a         a  c         c  =d         ?d  ee         ge  f         f  g         g  h         h  j         j  -k         /k  Ul         Wl  }m         m  n         n  o         o  p         p  r         r  Es         Gs  mt         ot  u         u  v         v  w         w  y         y  5z         7z  ]{         _{  |         |  }         }  ~         ~             %         '  M         O  u         w             Ņ         ǅ                        =         ?  e         g                        ݍ         ߍ             ,         .  S         U  {         }             ˔         ͔                        C         E  k         m                                              3         5  [         ]                        ӣ         գ             #         %  K         M  s         u             ê         Ū                        ;                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     y            #!          T!  1          1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    ^
         _
             2      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    `
         a
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            0     -,-X     ,-C-     C--    -<-x    <-J-|    J-}-\    }-]3-    ]3--h    -{b-    {b-n-	    n--h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     m            !          H!  1          1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             !         "  C         D  e         f                                               #  6          `  s                                              @  N          s                       
  "         #  ;          <   T!         U!  m"         n"  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            !         '  H         N  o         u                                              2	         8	  Y
         _
                                                       "  C         I  j         p                                              -         3  T         Z  {                                                       >          D   e!         k!  "          ),  F-          l-  .          .  /          /  1          51  R2         K  L         L  N         N  4O         >O  bP         lP  Q         Q  R         R  S         S  U         $U  HV         RV  vW         W  X         X  Y         Y  Z         [  &\         0\  R]         \]  ~^         ^  _         _  `         `  b         b  .c         8c  Xd         bd  e         e  f         f  g         g   i         
i  *j         4j  Tk         ^k  ~l         l  m         m  n         ȑ                        1         4  U         X  y         |                      ę             	           -         0  Q         T  u         x                                              )         ,  M         P  q         t                        ݪ                      %         (  I         L  m         p                        ٲ         ܲ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               L!         R!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     ?
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            8     -,-h     ,-y-     y--    -5-    5-C-    C-w-l    w---    ---x    -Kb-	    Kb--
    --
    -m-|    m-ճ-    ճ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ?!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        =  Z          z                                    1  /          N  L          k  i	          	  
          
                                   E         F  q         r           /  E         F  \         ]  s         t                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |        V"  |#         }#  $          B  C          C  D          D  E          F  0G          NG  bH         H  I         I  J         J  K         K  M         M  FN         GN  nO         oO  P         P  Q         Q  R         R  T         T  6U         7U  ^V         _V  W         W  X         X  Y         Y  Z         Z  &\         '\  N]         O]  v^         w^  _         _  `         `  a         a  c         c  >d         ?d  fe         ge  f         f  g         g  h         h  j         j  .k         /k  Vl         Wl  ~m         m  n         n  o         o  p         p  r         r  Fs         Gs  nt         ot  u         u  v         v  w         w  y         y  6z         7z  ^{         _{  |         |  }         }  ~         ~             &         '  N         O  v         w             ƅ         ǅ                        >         ?  f         g                        ލ         ߍ             -         .  T         U  |         }             ̔         ͔                        D         E  l         m                                              4         5  \         ]                        ԣ         գ             $         %  L         M  t         u             Ī         Ū                        <                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     x             !          P!  1          1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    ^
         _
             4      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    a
         b
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            0     -,-X     ,-a-     a--    -4-x    4-C-|    C-v-\    v--    --h    -3b-    3b-n-	    n--h
    -!׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               /!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        3  @          f  s                                              &  *          M  Q	          t	  x
          
                                              )         *  C                      4         5  K         L  b         c  y                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                     %  C         J  h         o                                              !	         (	  F
         M
  k         r                                              $         +  I         P  n         u                                            	  '         .  L         S  q         x                                                *!         1!  O"          /  0          0  1          $2  :3          `3  v4          4  5         I  J         J  K         K  L         M  !N         (N  FO         MO  kP         rP  Q         Q  R         R  S         S  T         U  $V         +V  IW         PW  nX         uX  Y         Y  Z         Z  [         [  ]         	]  '^         .^  L_         S_  q`         x`  a         a  b         b  c         c  	e         e  /f         6f  Ug         \g  {h         h  i         i  j         j  k         k  m         m  ;n         Dn  co         lo  p         p  q         q  r         r  t         t  +u         4u  Sv         \v  {w         w  x         x  y         y  z         z  |         $|  C}         L}  k~         t~                      Ā                        3         <  [         d                        ӈ         ܈             #         ,  K         T  s         |                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     l            !          D!  1          1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     @
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-:-x    :-oI-|    oI-|-\    |-W-    W--h    -]b-    ]b-x-	    x--h
    -!׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               J!         \!  1         
2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      ,         -  C	         D	  Z
         [
  q         5  N         O  h         i                                                                      ,  &          S  M          z                                        !          F!  S"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                        ?         @  _         `                                                        	  
          
  ?         @  _         `                                                                      ?         @  _         `                                                                      ?         @  _         `                          !         !  "         "  $         $  )%         +%  L&         N&  o'         q'  (         (  )         )  *         *  +         +  -          -  A.         C.  d/         f/  0         0  1         1  2         2  3         3  5         5  66         86  Y7         [7  |8         ~8  9         9  :         :  ;         ;  =         
=  +>         ->  N?         P?  q@         s@  A         A  B         B  C         "D  CE         EE  fF         hF  G         G  H         I  K         K  $L         &L  GM         IM  jN         lN  O         O  P         P  Q         Q  R         R  T         T  <U         >U  _V         aV  W         W  X         X  Y         Y  Z         Z  \         \  1]         3]  T^         V^  w_         y_  `         `  a         a  b         b  d         d  &e         (e  If         Kf  lg          Wl  Qm          m  yn          n  o          o  p          p  q                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     y            &!          Z!  1          2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-a-     a--    -4-x    4-C-|    C-v-\    v--    --h    -3b-    3b-n-	    n--h
    -!׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               1!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        3  A          f  t                                              &  )          M  P	          t	  w
          
                                              %         &  >                      /         0  F         G  ]         ^  t                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                     %  D         J  i         o                                              "	         (	  G
         M
  l         r                                               %         +  J         P  o         u                                            	  (         .  M         S  r         x                                                +!         1!  P"          5  6          7  *8          P8  f9          9  :          :  ;         I  J         J  K         K  L         M  "N         (N  GO         MO  lP         rP  Q         Q  R         R  S         S   U         U  %V         +V  JW         PW  oX         uX  Y         Y  Z         Z  [         [  ]         	]  (^         .^  M_         S_  r`         x`  a         a  b         b  c         c  
e         e  0f         6f  Vg         \g  |h         h  i         i  j         j  k         k  m         m  :n         Dn  bo         lo  p         p  q         q  r         r  t         t  *u         4u  Rv         \v  zw         w  x         x  y         y  z         z  |         $|  B}         L}  j~         t~                      Ā             
           2         <  Z         d                        ҈         ܈             "         ,  J         T  r         |                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     m            !          D!  1          1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     ?
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-:-x    :-iI-|    iI-|-\    |-!-    !--h    -]b-    ]b-x-	    x--h
    -!׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               F!         X!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      ,         -  C	         D	  Z
         [
  q         5  M         N  f         g                                                                       '             N  G          u                                       !          A!  O"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                        >         @  ^         `  ~                                                      	  
          
  >         @  ^         `  ~                                                                    >         @  ^         `  ~                                                                    >         @  ^         `  ~                        !         !  "         "  $         $  *%         +%  M&         N&  p'         q'  (         (  )         )  *         *  +         +  -          -  B.         C.  e/         f/  0         0  1         1  2         2  3         3  5         5  76         86  Z7         [7  }8         ~8  9         9  :         :  ;         ;  	=         
=  ,>         ->  O?         P?  r@         s@  A         A  B         B  C         "D  DE         EE  gF         hF  G         G  H         I  K         K  %L         &L  HM         IM  kN         lN  O         O  P         P  Q         Q  R         R  T         T  =U         >U  `V         aV  W         W  X         X  Y         Y  Z         Z  \         \  2]         3]  U^         V^  x_         y_  `         `  a         a  b         b  d         d  'e         (e  Jf         Kf  mg          r  s          Gs  At          ot  iu          u  v          v  w                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     x            %!          Z!  1          2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-m-     m--    -;-x    ;-I-|    I-}-\    }--    --h    -9b-    9b-x-	    x--h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               3!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         1  G         H  ^         _  u	         v	  
         
           ]  }         ~                                                           
          <  '          Y  D          v  a                                               J   \!          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                        <         @  \         `  |                                                      	  
          
  <         @  \         `  |                                                                    <         @  \         `  |                                                                    <         @  \         `  |                        !         !  "         "  $         $  $%         +%  G&         N&  j'         q'  (         (  )         )  *         *  +         +  -          -  <.         C.  _/         f/  0         0  1         1  2         2  3         3  5         5  16         86  T7         [7  w8         ~8  9         9  :         :  ;         ;  =         
=  &>         ->  I?         P?  l@         s@  A         A  B         B  C         "D  >E         EE  aF         hF  G         G  H         I  J         K  L         &L  BM         IM  eN         lN  O         O  P         P  Q         Q  R         R  T         T  7U         >U  ZV         aV  }W         W  X         X  Y         Y  Z         Z  	\         \  ,]         3]  O^         V^  r_         y_  `         `  a         a  b         b  c         d  !e         (e  Df         Kf  gg            	          <  1          d  Y                                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     l            !          F!  1          1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 y
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-#5-x    #5-C-|    C-w-\    w-
-    
--h    -ub-    ub-x-	    x--h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               L!         \!  1         
2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        =  Y          z                                    1  0          N  M          k  j	          	  
          
                                   B         C  m         n           *  @         A  W         X  n         o                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                 (  5          V  c                     e$  %         %  &         &  '         '  (         (  *         
*  )+         ++  J,         L,  k-         m-  .         .  /         /  0         0  1         1  3         3  14         34  R5         T5  s6         u6  7         7  8         8  9         9  :         :  <         <  9=         ;=  Z>         \>  {?         }?  @         @  A         A  B         B  C         D   E         "E  AF         F  G         G  I         
I  )J         +J  JK         LK  kL         mL  M         M  N         N  O         O  P         P  R         R  1S         3S  RT         TT  sU         uU  V         V  W         W  X         X  Y         Y  [         [  9\         ;\  Z]         \]  {^         }^  _         _  `         `  a         a  b         c   d         "d  Ae         Ce  bf         df  g         g  h         h   j         "j  Dk         Fk  hl         jl  m         m  n         n  o         o  p         p  r         r  @s         Bs  dt         ft  u         u  v         v  w         w  x         x  z         z  <{         >{  `|         b|  }         }  ~         ~                                   8         :  \         ^                        ȇ         ʇ                        4                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               @!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-g-     g--    -);-x    );-I-|    I-}-\    }--    --h    -?b-    ?b-x-	    x--h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               1!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         1  G         H  ^         _  u	         v	  
         
           b                                                                    )            F  0          c  M            j                                    *           T   g!          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                        =         @  ]         `  }                                                      	  
          
  =         @  ]         `  }                                                                    =         @  ]         `  }                                                                    =         @  ]         `  }                        !         !  "         "  $         $  %%         +%  H&         N&  k'         q'  (         (  )         )  *         *  +         +  -          -  =.         C.  `/         f/  0         0  1         1  2         2  3         3  5         5  26         86  U7         [7  x8         ~8  9         9  :         :  ;         ;  =         
=  '>         ->  J?         P?  m@         s@  A         A  B         B  C         "D  ?E         EE  bF         hF  G         G  H         I  J         K   L         &L  CM         IM  fN         lN  O         O  P         P  Q         Q  R         R  T         T  8U         >U  [V         aV  ~W         W  X         X  Y         Y  Z         Z  
\         \  -]         3]  P^         V^  s_         y_  `         `  a         a  b         b  c         d  "e         (e  Ef         Kf  hg          ܈  щ                      ,  !          T  I          |  q                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     m            !          J!  1          1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 y
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-)5-x    )5-C-|    C-w-\    w-
-    
--h    -ub-    ub-x-	    x--h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               H!         X!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        =  Z          z                                    1  /          N  L          k  i	          	  
          
                                   E         F  q         r           /  E         F  \         ]  s         t                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                             <  I          j  w         e$  %         %  &         &  '         '  (         (  	*         
*  *+         ++  K,         L,  l-         m-  .         .  /         /  0         0  1         1  3         3  24         34  S5         T5  t6         u6  7         7  8         8  9         9  :         :  <         <  :=         ;=  [>         \>  |?         }?  @         @  A         A  B         B   D         D  !E         "E  BF         F  G         G  	I         
I  *J         +J  KK         LK  lL         mL  M         M  N         N  O         O  P         P  R         R  2S         3S  ST         TT  tU         uU  V         V  W         W  X         X  Y         Y  [         [  :\         ;\  []         \]  |^         }^  _         _  `         `  a         a   c         c  !d         "d  Be         Ce  cf         df  g         g  h         h  !j         "j  Ek         Fk  il         jl  m         m  n         n  o         o  p         p  r         r  As         Bs  et         ft  u         u  v         v  w         w  x         x  z         z  ={         >{  a|         b|  }         }  ~         ~                                   9         :  ]         ^                        ɇ         ʇ                        5                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               @!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-s-     s--    -5-x    5-C-|    C-v-\    v-
-    
--h    -Kb-    Kb-x-	    x--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               5!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        3  @          f  s                                              &  *          M  Q	          t	  x
          
                                              )         *  C                      4         5  K         L  b         c  y                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       ,  4          X  `                                 e$  %         %  &         &  '         '  (         (  *         
*  $+         ++  E,         L,  f-         m-  .         .  /         /  0         0  1         1  3         3  ,4         34  M5         T5  n6         u6  7         7  8         8  9         9  :         :  <         <  4=         ;=  U>         \>  v?         }?  @         @  A         A  B         B  C         D  E         "E  <F         F  G         G  I         
I  $J         +J  EK         LK  fL         mL  M         M  N         N  O         O  P         P  R         R  ,S         3S  MT         TT  nU         uU  V         V  W         W  X         X  Y         Y  [         [  4\         ;\  U]         \]  v^         }^  _         _  `         `  a         a  b         c  d         "d  <e         Ce  ]f         df  ~g         g  h         h  j         "j  =k         Fk  al         jl  m         m  n         n  o         o  p         p  r         r  9s         Bs  ]t         ft  u         u  v         v  w         w  x         x  z         z  5{         >{  Y|         b|  }}         }  ~         ~                                   1         :  U         ^  y                               ʇ             	           -                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               )!         8!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 y
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-:-x    :-uI-|    uI-|-\    |-+-    +--h    -]b-    ]b--	    --h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               D!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      ,         -  C	         D	  Z
         [
  q         5  N         O  h         i                                                                      ,  &          S  M          z                                        !          F!  S"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       7         8  S         T  o         p                                              	         	             3         4  O         P  k         l                                                                    /         0  K         L  g         h                                                                    +          ,   G!         !  "         "  #         #  %         %  :&         <&  Y'         ['  x(         z(  )         )  *         *  +         +  ,         ,  .         .  2/         4/  Q0         S0  p1         r1  2         2  3         3  4         4  5         5  7         7  *8         ,8  I9         K9  h:         j:  ;         ;  <         <  =         =  >         >  @         @  "A         $A  AB         CB  `C         C  E         E  4F         6F  SG         UG  rH         tH  I          \  d                                                         (  E         G  d         f                                 Õ                                    =         ?  \         ^  {         }                        ؟         ڟ                        5         7  T         V  s         u                        Ш         Ҩ                        -                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               /!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   K      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-m-     m--    -4-x    4-C-|    C-v-\    v-
-    
--h    -Qb-    Qb-x-	    x--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               3!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        3  A          f  t                                              &  )          M  P	          t	  w
          
                                              %         &  >                      /         0  F         G  ]         ^  t                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                 4  <	          `	  h
          
           e$  %         %  &         &  '         '  (         (  *         
*  %+         ++  F,         L,  g-         m-  .         .  /         /  0         0  1         1  3         3  -4         34  N5         T5  o6         u6  7         7  8         8  9         9  :         :  <         <  5=         ;=  V>         \>  w?         }?  @         @  A         A  B         B  C         D  E         "E  =F         F  G         G  I         
I  %J         +J  FK         LK  gL         mL  M         M  N         N  O         O  P         P  R         R  -S         3S  NT         TT  oU         uU  V         V  W         W  X         X  Y         Y  [         [  5\         ;\  V]         \]  w^         }^  _         _  `         `  a         a  b         c  d         "d  =e         Ce  ^f         df  g         g  h         h  j         "j  <k         Fk  `l         jl  m         m  n         n  o         o  p         p  r         r  8s         Bs  \t         ft  u         u  v         v  w         w  x         x  z         z  4{         >{  X|         b|  |}         }  ~         ~                                   0         :  T         ^  x                               ʇ                        ,                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               -!         <!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 y
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-:-x    :-uI-|    uI-|-\    |-+-    +--h    -]b-    ]b--	    --h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               D!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      ,         -  C	         D	  Z
         [
  q         5  M         N  f         g                                                                       '             N  G          u                                       !          A!  O"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       6         8  R         T  n         p                                              	         	             2         4  N         P  j         l                                                                    .         0  J         L  f         h                                                                    *          ,   F!         !  "         "  #         #  %         %  ;&         <&  Z'         ['  y(         z(  )         )  *         *  +         +  ,         ,  .         .  3/         4/  R0         S0  q1         r1  2         2  3         3  4         4  5         5  7         7  +8         ,8  J9         K9  i:         j:  ;         ;  <         <  =         =  >         >  @         @  #A         $A  BB         CB  aC         C  E         E  5F         6F  TG         UG  sH         tH  I          B  J          p  x                      ̍  Ԏ                     (  F         G  e         f                                 Õ                                     >         ?  ]         ^  |         }                        ٟ         ڟ                        6         7  U         V  t         u                        Ѩ         Ҩ                        .                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               /!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   L      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-O-     O--    -);-x    );-I-|    I-}-\    }-+-    +--h    -]b-    ]b-3-	    3--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               $!         4!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         1  G         H  ^         _  u	         v	  
         
           ]  }         ~                                                           
          <  '          Y  D          v  a                                               J   \!          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       4         8  P         T  l         p                                              	         	             0         4  L         P  h         l                                                                    ,         0  H         L  d         h                                                                    (          ,   D!         !  "         "  #         #  %         %  5&         <&  T'         ['  s(         z(  )         )  *         *  +         +  ,         ,  .         .  -/         4/  L0         S0  k1         r1  2         2  3         3  4         4  5         5  7         7  %8         ,8  D9         K9  c:         j:  ;         ;  <         <  =         =  >         >  ?         @  A         $A  <B         CB  [C         C  E         E  /F         6F  NG         UG  mH         tH  I          V  +W          SW  kX          X  Y          Y  Z          [  +\         (  @         G  _         f  ~                               Õ  ۖ                                  8         ?  W         ^  v         }                        ӟ         ڟ                        0         7  O         V  n         u                        ˨         Ҩ             	           (                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ,!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
  d      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-I<-x    I<-J-|    J-1~-\    1~-S-    S--h    -Eb-    Eb--	    --h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               D!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        3         4  M         N  g         h                                      &          L  Y                                              ,  H          i                                    7          8   b!         c!  "         "  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       :         <  X         Z  v         x                                              
         
  *         ,  H         J  f         h                                                                    8         :  V         X  t         v                                              
           (         *  F         H  d          f   !         #  $         $  &         &  :'         <'  Z(         \(  z)         |)  *         *  +         +  ,         ,  -         -  /         /  :0         <0  Z1         \1  z2         |2  3         3  4         4  5         5  6         6  8         8  :9         <9  Z:         \:  z;         |;  <         <  =         =  >         >  ?         ?  A         A  :B         <B  ZC         \C  zD         |D  E         F  G         G  H         H  J         J  ?K         @K  eL         fL  M         M  N         N  O         O  P         P  #R         $R  ES         GS  hT         jT  U         U  V         V  W         W  X         X  Z         Z  :[         <[  ]\         _\  ]         ]  ^         ^  _         _  `         `  b         b  %c         'c  Fd         Hd  ge         ie  f         f  g         g  h          i  j          j  l          <l  Ym          m  n          n  o                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               (!         2!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   K      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-O-     O--    -5;-x    5;-I-|    I-}-\    }-+-    +--h    -Wb-    Wb-9-	    9--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               #!         2!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         1  G         H  ^         _  u	         v	  
         
           b                                                                    )            F  0          c  M            j                                    *           T   g!          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       5         8  Q         T  m         p                                              	         	             1         4  M         P  i         l                                                                    -         0  I         L  e         h                                                                    )          ,   E!         !  "         "  #         #  %         %  6&         <&  U'         ['  t(         z(  )         )  *         *  +         +  ,         ,  .         .  ./         4/  M0         S0  l1         r1  2         2  3         3  4         4  5         5  7         7  &8         ,8  E9         K9  d:         j:  ;         ;  <         <  =         =  >         >  ?         @  A         $A  =B         CB  \C         C  E         E  0F         6F  OG         UG  nH         tH  I          S\  k]          ]  ^          ^  _          `  +a          Sa  kb         (  A         G  `         f                                 Õ  ܖ                                  9         ?  X         ^  w         }                        ԟ         ڟ                        1         7  P         V  o         u                        ̨         Ҩ             
           )                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               *!         @!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
  e      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-I<-x    I<-J-|    J-1~-\    1~--    --h    -Kb-    Kb--	    --h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               @!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        1         2  J         K  c         d  |                                    "          G  U          z                                    '  D          d                                  	  4          5   `!         a!  "         "  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       ;         <  Y         Z  w         x                                              
         
  +         ,  I         J  g         h                                                                    9         :  W         X  u         v                                                         )         *  G         H  e          f   !         #  $         $  &         &  ;'         <'  [(         \(  {)         |)  *         *  +         +  ,         ,  -         -  /         /  ;0         <0  [1         \1  {2         |2  3         3  4         4  5         5  6         6  8         8  ;9         <9  [:         \:  {;         |;  <         <  =         =  >         >  ?         ?  A         A  ;B         <B  [C         \C  {D         |D  E         F  G         G  H         H  J         J  >K         @K  dL         fL  M         M  N         N  O         O  P         P  "R         $R  FS         GS  iT         jT  U         U  V         V  W         W  X         X  Z         Z  ;[         <[  ^\         _\  ]         ]  ^         ^  _         _  `         `  b         b  &c         'c  Gd         Hd  he         ie  f         f  g         g  h          p  "q          Hq  er          r  s          s  t          u  .v                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               *!         6!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   L      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-<-x    <-J-|    J-~-\    ~--    --h    -ib-    ib-9-	    9--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               =!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                       !  A         B  b         c                                                 0          [  m                                              ;  H          n  {                                  8          9   R!         S!  l"         m"  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       5         <  S         Z  q         x                                              
         
  %         ,  C         J  a         h                                                                    3         :  Q         X  o         v                                                         #         *  A         H  _          f   }!         #  $         $  &         &  3'         <'  S(         \(  s)         |)  *         *  +         +  ,         ,  -         -  /         /  30         <0  S1         \1  s2         |2  3         3  4         4  5         5  6         6  8         8  39         <9  S:         \:  s;         |;  <         <  =         =  >         >  ?         ?  A         A  3B         <B  SC         \C  sD         |D  E         F  G         G  H         H  J         J  <K         @K  bL         fL  M         M  N         N  O         O  P         P   R         $R  @S         GS  cT         jT  U         U  V         V  W         W  X         X  Z         Z  5[         <[  X\         _\  {]         ]  ^         ^  _         _  `         `  a         b   c         'c  Ad         Hd  be         ie  f         f  g         g  h                        Ă                        $          L  T                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               1!         J!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
  d      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-:-x    :-I-|    I-|-\    |-+-    +--P    -{b-    {b-c-	    c-g-D
    g-3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               A!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      ,         -  C	         D	  Z
         [
  q         5  N         O  h         i                                                                      ,  &          S  M          z                                        !          F!  S"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p                        ?         @  _         `                                                        	  
          
  ?         @  _         `                                                                      ?         @  _         `                                                                      ?         @  _         `                          !         !  "         "  #          $  %          %  >&         @&  ^'         `'  ~(         (  )         )  *         *  +         +  ,          -  .          .  >/         @/  ^0         `0  ~1         1  2         2  3         3  4         4  5          6  7          7  >8         @8  ^9         `9  ~:         :  ;         ;  <         <  =         =  >          ?  @          @  >A         G  
I         I  *J         +J  JK         KK  jL         kL  M         M  N         N  O         O  P         P  
R         R  *S         +S  JT         KT  jU         kU  V         V  W         W  X         X  Y         Y  
[         [  *\         +\  J]         K]  j^         k^  _         _  `         `  a         a  b         b  d         d  6e         7e  Xf         Yf  zg         {g  h          y  z          {  !|          D|  S}          v}  ~          ~                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               =!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 }
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-<-x    <-J-|    J-~-\    ~--    --h    -ib-    ib-9-	    9--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               A!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             !         "  C         D  e         f                                               #  6          `  s                                              @  N          s                       
  "         #  ;          <   T!         U!  m"         n"  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       6         <  T         Z  r         x                                              
         
  &         ,  D         J  b         h                                                                    4         :  R         X  p         v                                                         $         *  B         H  `          f   ~!         #  $         $  &         &  2'         <'  R(         \(  r)         |)  *         *  +         +  ,         ,  -         -  /         /  20         <0  R1         \1  r2         |2  3         3  4         4  5         5  6         6  8         8  29         <9  R:         \:  r;         |;  <         <  =         =  >         >  ?         ?  A         A  2B         <B  RC         \C  rD         |D  E         F  G         G  H         H  J         J  =K         @K  cL         fL  M         M  N         N  O         O  P         P  !R         $R  AS         GS  dT         jT  U         U  V         V  W         W  X         X  Z         Z  6[         <[  Y\         _\  |]         ]  ^         ^  _         _  `         `   b         b  !c         'c  Bd         Hd  ce         ie  f         f  g         g  h          |                        ܈                        <  D                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               1!         J!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
  e      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-:-x    :-{I-|    {I-|-\    |-%-    %--P    -{b-    {b-c-	    c-m-D
    m-3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               >!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      ,         -  C	         D	  Z
         [
  q         5  M         N  f         g                                                                       '             N  G          u                                       !          A!  O"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p                        >         @  ^         `  ~                                                      	  
          
  >         @  ^         `  ~                                                                    >         @  ^         `  ~                                                                    >         @  ^         `  ~                        !         !  "         "  #          $  %          %  ?&         @&  _'         `'  (         (  )         )  *         *  +         +  ,          -  .          .  ?/         @/  _0         `0  1         1  2         2  3         3  4         4  5          6  7          7  ?8         @8  _9         `9  :         :  ;         ;  <         <  =         =  >          ?  @          @  ?A         G  	I         I  )J         +J  IK         KK  iL         kL  M         M  N         N  O         O  P         P  	R         R  )S         +S  IT         KT  iU         kU  V         V  W         W  X         X  Y         Y  	[         [  )\         +\  I]         K]  i^         k^  _         _  `         `  a         a  b         b  d         d  5e         7e  Wf         Yf  yg         {g  h                                  >  M          p                                 >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               =!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 }
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    <     -,-p     ,--     -)-    )-55-    55-C-    C-w-t    w-12-     12-9-X    9--    -b-p	    b--x
    --
    -s-    s-ճ-\    ճ--<                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ?!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        3  @          f  s                                               &  '          M  N	          t	  u
          
                                              )         *  C           $         %  B         C  `         a  ~                                         b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h         $  %         K  L         L  N         N  EO         FO  mP         nP  Q         Q  R         R  S         S  U         U  5V         6V  ]W         ^W  X         X  Y         Y  Z         Z  [         [  %]         &]  M^         N^  u_         v_  `         `  a         a  b         b  d         d  =e         >e  ef         ff  g         g  h         h  i         i  k         k  -l         .l  Um         Vm  }n         ~n  o         o  p         p  q         q  s         s  Et         Ft  mu         nu  v         v  w         w  x         x  z         z  5{         6{  ]|         ^|  }         }  ~         ~                        %         &  M         N  u         v             ņ         Ɔ                        =         >  e         f                        ݎ         ގ             -          .  3          V  [          ~                       Ε                        D         F  l         n                                              4         6  \         ^                        Ԥ         ֤             $         &  L         N  t         v             ī         ƫ                        <         >  d         f                                           b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               @!         P!  1         1  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    ^
         _
             &                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    ^
         _
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     t
            8     -,-h     ,--     -#-    #-y<-    y<-K-    K-a~-l    a~--    -G-P    G--    -b-h	    b-!-p
    !-/-
    /-簳-    簳-9׳-L                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               >!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        3         4  M         N  g         h                                      &          L  Y                                              ,  C          i                                    I          J   }!         ~!  "         "  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        $  8%         :%  [&         \&  }'         ~'  (         (  )         )  *         *  ,         ,  '-         (-  I.         J.  k/         l/  0         0  1         1  2         2  3         3  5         5  76         86  Y7         Z7  {8         |8  9         9  :         :  ;         ;  =         =  %>         &>  G?         H?  i@         j@  A         A  B         B  C         C  D         D  F         F  5G         6G  VH         XH  xI         zI  J         J  K         K  L         L  "N         $N  UO         WO  P         P  Q         Q  R         R  !T         #T  TU         VU  V         V  W         W  X         X  'Z         (Z  ][         ^[  \         \  ]         ]  ^          _  5`         6`  ka         la  b         b  c         c  e         e  ?f         @f  sg         tg  h         h  i         i  k         k  Cl         Dl  wm         xm  n         n  o         o   q         q  "r         $r  Ds         Fs  ft         ht  u         u  v         v  w         w  x         x  z         z  2{         4{  T|         V|  v}         x}  ~         ~             ܀          ހ  ݁                       "  !          D  C          f  e                      ʈ         ̈                        0                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               J!         ^!  1         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    ^
         _
             F                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     C
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    <     -,-p     ,--     -5-    5-)5-    )5-C-    C-w-t    w-12-     12-9-X    9--    -b-p	    b--x
    --
    -s-    s-ճ-\    ճ--<                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               C!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        3  A          f  t                                              &  (          M  O	          t	  v
          
                                              %         &  >                      ;         <  X         Y  u         v                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        $  %         K  L         L  N         N  DO         FO  lP         nP  Q         Q  R         R  S         S  U         U  4V         6V  \W         ^W  X         X  Y         Y  Z         Z  [         [  $]         &]  L^         N^  t_         v_  `         `  a         a  b         b  d         d  <e         >e  df         ff  g         g  h         h  i         i  k         k  ,l         .l  Tm         Vm  |n         ~n  o         o  p         p  q         q  s         s  Dt         Ft  lu         nu  v         v  w         w  x         x  z         z  4{         6{  \|         ^|  }         }  ~         ~                        $         &  L         N  t         v             Ć         Ɔ                        <         >  d         f                        ܎         ގ             ,         .  U         V  }         ~             ͕          Ε  Ӗ                        #          F  K          n  s                                            5         6  ]         ^                        դ         ֤             %         &  M         N  u         v             ū         ƫ                        =         >  e         f                                           b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               B!         P!  1         1  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    ^
         _
             &                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    ^
         _
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     s
            8     -,-h     ,--     -#-    #-s<-    s<-K-    K-[~-l    [~--    -)-P    )--    -b-h	    b-!-p
    !-/-
    /-簳-    簳-9׳-L                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               <!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        1         2  J         K  c         d  |                                    "          G  U          z                                    '  ?          d  |                                  B          C   u!         v!  "         "  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        $  9%         :%  Z&         \&  |'         ~'  (         (  )         )  *         *  ,         ,  &-         (-  H.         J.  j/         l/  0         0  1         1  2         2  3         3  5         5  66         86  X7         Z7  z8         |8  9         9  :         :  ;         ;  =         =  $>         &>  F?         H?  h@         j@  A         A  B         B  C         C  D         D  F         F  4G         6G  WH         XH  yI         zI  J         J  K         K  L         L  #N         $N  VO         WO  P         P  Q         Q  R         R  "T         #T  UU         VU  V         V  W         W  X         X  &Z         (Z  \[         ^[  \         \  ]         ]  ^          _  4`         6`  ja         la  b         b  c         c  
e         e  >f         @f  rg         tg  h         h  i         i  k         k  Bl         Dl  vm         xm  n         n  o         o  q         q  #r         $r  Es         Fs  gt         ht  u         u  v         v  w         w  x         x  z         z  3{         4{  U|         V|  w}         x}  ~         ~             ݀         ހ              !         "  C         D  e         f                                    ̈  ˉ                                                      b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               H!         ^!  1         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    ^
         _
             F                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     D
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -G-    G-1<-x    1<-J-|    J-~-\    ~-4-    4--h    -b-    b-n-	    n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               O!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             #         $  G         H  k         l                                               -  4          j  q                                    $          J  W          }                         -         .  G          H   a!         b!  {"         |"  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            $         &  J         L  p         r                                            
  .	         0	  T
         V
  z         |                                              8         :  ^         `                                                         B         D  h         j                                               &          (   L!         N!  r"          0  1          1  3          03  E4          h4  }5          5  6         K  L         L  M          N  %O         &O  KP         LP  qQ         rQ  R         R  S         S  T         T  	V         
V  /W         0W  UX         VX  {Y         |Y  Z         Z  [         [  \         \  ^         ^  9_         :_  _`         ``  a         a  b         b  c         c  d         d  f         f  Cg         Dg  ih         jh  i         i  j         j  k         k  m         m  'n         (n  Mo           ?         @  e         f                        י         ؙ             #         $  I         J  o         p                                              -         .  S         T  y         z             Ũ         ƨ                        7         8  ]         ^                        ϰ         а                        A         B  g                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               P!         V!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     5
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     -;-    ;-/5-    /5-C-    C-w-l    w-5-    5--x    -b-	    b-	-
    	--
    -y-|    y-ճ-    ճ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               I!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        =  Q          z                                    1             N  =          k  Z	          	  w
          
                                   *         +  M         N  p            6         7  M         N  d         e  {         |                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |        8"  Y#         [#  |$         ~$  %         %  &         &  '         '  )         )  0*         2*  T+         V+  x,         z,  -         -  .         .  /         /  1         
1  ,2         .2  P3         R3  t4         v4  5         5  6         6  7         7  9         9  (:         *:  L;         N;  p<         r<  =         =  >         >  ?         ?   A         A  $B         &B  HC         JC  lD         nD  E         E  F         F  G          G  H          I  J          EJ  UK          {K  L          L  M         -l  Nm         Pm  qn         sn  o         o  p         p  q         q  s         s  %t         't  Iu         Ku  mv         ov  w         w  x         x  y         y  z         z  !|         #|  E}         G}  i~         k~                        Ձ         ׁ                        A         C  e         g                        щ         Ӊ           ړ                       "  D         F  h         j                        ԛ         ֛                        @         B  d         f                        У         ң                        <         >  `         b                        ̫         Ϋ                        8         :  \         ^                        ȳ         ʳ                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               H!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    ^
         _
             (      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    ^
         _
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     c
            0     -,-X     ,--     -G-    G-+<-x    +<-J-|    J-~-\    ~-4-    4--h    -b-    b-n-	    n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               O!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             "         #  E         F  h         i                                               (  0          e  m                                               E  S          x                         '         (  @          A   Y!         Z!  r"         s"  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            %         &  K         L  q         r                                   	         
  /	         0	  U
         V
  {         |                                              9         :  _         `                                                         C         D  i         j                                              '          (   M!         N!  s"          6  7          8  %9          H9  ]:          :  ;          ;  <         K  L         L  M          N  $O         &O  JP         LP  pQ         rQ  R         R  S         S  T         T  V         
V  .W         0W  TX         VX  zY         |Y  Z         Z  [         [  \         \  ^         ^  8_         :_  ^`         ``  a         a  b         b  c         c  d         d  f         f  Bg         Dg  hh         jh  i         i  j         j  k         k   m         m  &n         (n  Lo           >         @  d         f                        ֙         ؙ             "         $  H         J  n         p                                              ,         .  R         T  x         z             Ĩ         ƨ                        6         8  \         ^                        ΰ         а                        @         B  f                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               R!         Z!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     4
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     -;-    ;-;5-    ;5-C-    C-#w-l    #w-5-    5--x    -b-	    b-	-
    	--
    -y-|    y-ճ-    ճ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               I!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        =  R          z                          	          1            N  <          k  Y	          	  v
          
                        	         
  -         .  Q         R  u         *  @         A  W         X  n         o                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |        8"  Z#         [#  }$         ~$  %         %  &         &  '         '  )         )  1*         2*  U+         V+  y,         z,  -         -  .         .  /         /  	1         
1  -2         .2  Q3         R3  u4         v4  5         5  6         6  7         7  9         9  ):         *:  M;         N;  q<         r<  =         =  >         >  ?         ?  A         A  %B         &B  IC         JC  mD         nD  E         E  F         F  G          M  N          O  -P          SP  cQ          Q  R          R  S         -l  Om         Pm  rn         sn  o         o  p         p  q         q  s         s  &t         't  Ju         Ku  nv         ov  w         w  x         x  y         y  z         z  "|         #|  F}         G}  j~         k~                        ց         ׁ                        B         C  f         g                        ҉         Ӊ           ړ             !         "  E         F  i         j                        ՛         ֛                        A         B  e         f                        ѣ         ң                        =         >  a         b                        ͫ         Ϋ                        9         :  ]         ^                        ɳ         ʳ                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               F!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    ^
         _
             )      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    ^
         _
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     d
            0     -,-X     ,--     -;-    ;-5-x    5-C-|    C-v-\    v--    --h    -b-    b-'n-	    'n--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               K!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        3  @          f  s                                              &  "          M  I	          t	  p
          
                                              )         *  C                      4         5  K         L  b         c  y                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            "         $  F         H  j         l                                              	          	  B
         D
  f         h                                                         >         @  b         d                                                         :         <  ^         `                                              !         !  6"          <  =          =  >          ?  3@          Y@  vA          A  B         .I  PJ         RJ  tK         vK  L         L  M         M  N         N  P         P  (Q         *Q  LR         NR  pS         rS  T         T  U         U  V         V   X         X  $Y         &Y  HZ         JZ  l[         n[  \         \  ]         ]  ^         ^  _         _  #a         $a  Ib         Jb  oc         pc  d         d  e         e  f         f  h         h  -i         .i  Sj         Tj  yk         m  n         n  p         p  <q         =q  `r         ar  s         s  t         t  u         u  v         v  x         x  8y         9y  \z         ]z  {         {  |         |  }         }  ~         ~             4         5  X         Y  |         }             ą         Ņ                        0         1  T         U  x         y                                                      >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               G!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     H
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -;-    ;-+<-x    +<-J-|    J-~-\    ~--    --h    -b-    b--n-	    -n--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               H!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        3         4  M         N  g         h                                      &          L  Y                                              ,  @          i  }                                  '          (   J!         K!  m"         n"  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            $         &  J         L  p         r                                            
  .	         0	  T
         V
  z         |                                              8         :  ^         `                                                         B         D  h         j                                               &          (   L!         N!  r"         t"  #         #  $         $  %         %  '         '  1(         2(  W)         X)  }*         ~*  +         +  ,         ,  -         -  /         /  ;0         <0  a1         b1  2          &?  @@          i@  A          A  B          B  	D          2D  LE         _j  k         k  l         l  m         m  n         n  p         p  Aq         Cq  gr         ir  s         s  t         t  u         u  v         w  %x         'x  Ky         My  qz         sz  {         {  |         |  }         }  	           /         1  U         W  {         }             Ǆ         Ʉ                        9         ;  _         a                        ь         ӌ                      	  .         /  T         U  z         {             ƕ         Ǖ                        8         9  ^         _                        Н         ѝ                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               I!         V!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     O
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -;-    ;-5-x    5-C-|    C-v-\    v--    --h    -b-    b-'n-	    'n--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               K!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        3  A          f  t                                              &  !          M  H	          t	  o
          
                                              %         &  >                      /         0  F         G  ]         ^  t                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            #         $  G         H  k         l                                              	          	  C
         D
  g         h                                                         ?         @  c         d                                                         ;         <  _         `                                              !         !  7"          B  C          "D  ?E          eE  F          F  G          G  I         .I  QJ         RJ  uK         vK  L         L  M         M  N         N  P         P  )Q         *Q  MR         NR  qS         rS  T         T  U         U  V         V  X         X  %Y         &Y  IZ         JZ  m[         n[  \         \  ]         ]  ^         ^  _         _  "a         $a  Hb         Jb  nc         pc  d         d  e         e  f         f  h         h  ,i         .i  Rj         Tj  xk         m  n         n  p         p  ;q         =q  _r         ar  s         s  t         t  u         u  v         v  x         x  7y         9y  [z         ]z  {         {  |         |  }         }  ~         ~             3         5  W         Y  {         }             Å         Ņ                        /         1  S         U  w         y                                                      >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               G!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     G
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -;-    ;-+<-x    +<-J-|    J-~-\    ~--    --h    -b-    b--n-	    -n--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               J!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        1         2  J         K  c         d  |                                    "          G  U          z                                    '  <          d  y                                   $          %   H!         I!  l"         m"  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            %         &  K         L  q         r                                   	         
  /	         0	  U
         V
  {         |                                              9         :  _         `                                                         C         D  i         j                                              '          (   M!         N!  s"         t"  #         #  $         $  %         %  
'         '  0(         2(  V)         X)  |*         ~*  +         +  ,         ,  -         -  /         /  :0         <0  `1         b1  2          uE  F          F  G          G  I          >I  XJ          J  K         _j  k         k  l         l  m         m  n         n  p         p  Bq         Cq  hr         ir  s         s  t         t  u         u   w         w  &x         'x  Ly         My  rz         sz  {         {  |         |  }         }  
           0         1  V         W  |         }             Ȅ         Ʉ                        :         ;  `         a                        Ҍ         ӌ                      	  -         /  S         U  y         {             ŕ         Ǖ                        7         9  ]         _                        ϝ         ѝ                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               I!         V!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     N
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-5;-x    5;-I-|    I-}-\    }-+-    +--h    -b-    b-x-	    x--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               I!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         1  G         H  ^         _  u	         v	  
         
           5  ]         ^                                                                     <  7          Y  T          v  q                                    '           J   d!          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                      "  B         D  d         f                                              	         	  0
         2
  R         T  t         v                                                          @         B  b         d                                                         .         0  P         R  r         t                                     !         "  1#         3#  S$         sG  H         H  I         I  J         J  K         K  M         M  =N         ?N  _O         aO  P         P  Q         Q  R         R  S         S  	U         U  +V         -V  MW         OW  oX         qX  Y         Y  Z         Z  [         [  \         \  ^         ^  ;_         =_  ]`         _`  a         a  b         b  c         c  d         d  f         	f  )g         +g  Kh         Mh  mi         oi  j         j  k         k  l          m  n          n  o          p  q          =q  >r          ar  bs         $  D         F  f         h                        ̕         Ε                        2         4  T         V  v         x                        ܞ         ޞ                        "  B         D  d         f                        ʦ         ̦                        0         2  R         T  t                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               B!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 y
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-)5-x    )5-C-|    C-w-\    w-C-    C--h    -b-    b-'n-	    'n--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               H!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        =  Q          z                                    1             N  =          k  Z	          	  w
          
                                   *         +  M         N  p            6         7  M         N  d         e  {         |                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            &         (  N         P  v         x                                              >	         @	  f
         h
                                              .         0  V         X  ~                                                        F         H  n         p                                              6         8  ^          `   !         !  "         "  $         $  :%         <%  b&         d&  '         '  (         (  )         )  +         +  *,         ,,  R-         T-  z.         |.  /         /  0         0  1         1  3         3  B4         D4  j5         l5  6         6  7         7  8         8  
:         :  2;         4;  Z<         \<  =         =  >         >  ?         ?  @         @  "B         $B  JC         LC  rD         tD  E         E  F          `  a          &b  9c          bc  ud          d  e          e  f         Bm  hn         jn  o         o  p         p  q         q  s         
s  0t         2t  Xu         Zu  v         v  w         w  x         x  y         y   {         "{  H|         J|  p}         r}  ~         ~                                   8         :  `         b                        ؇         ڇ              (         *  P         R  x         z             Ȏ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               H!         T!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     O
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-A;-x    A;-I-|    I-)}-\    )}-+-    +--h    -b-    b-x-	    x--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               E!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         1  G         H  ^         _  u	         v	  
         
           :  c         d                                                         )  #          F  @          c  ]            z                                    2           T   o!          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            !         "  C         D  e         f                                              	         	  1
         2
  S         T  u         v                                                          A         B  c         d                                                         /         0  Q         R  s         t                                     !         "  2#         3#  T$         sG  H         H  I         I  J         J  K         K  M         M  >N         ?N  `O         aO  P         P  Q         Q  R         R  S         S  
U         U  ,V         -V  NW         OW  pX         qX  Y         Y  Z         Z  [         [  \         \  ^         ^  <_         =_  ^`         _`  a         a  b         b  c         c  d         d  f         	f  *g         +g  Lh         Mh  ni         oi  j         j  k         k  l          s  t          t  u          u  v          v  w          x  y         $  E         F  g         h                        ͕         Ε                        3         4  U         V  w         x                        ݞ         ޞ              !         "  C         D  e         f                        ˦         ̦                        1         2  S         T  u                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               D!         R!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 y
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -;-    ;-55-x    55-C-|    C-w-\    w-C-    C--h    -b-    b-'n-	    'n--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               J!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        =  R          z                          	          1            N  <          k  Y	          	  v
          
                        	         
  -         .  Q         R  u         *  @         A  W         X  n         o                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            '         (  O         P  w         x                                              ?	         @	  g
         h
                                              /         0  W         X                                                          G         H  o         p                                              7         8  _          `   !         !  "         "  $         $  ;%         <%  c&         d&  '         '  (         (  )         )  +         +  +,         ,,  S-         T-  {.         |.  /         /  0         0  1         1  3         3  C4         D4  k5         l5  6         6  7         7  8         8  :         :  3;         4;  [<         \<  =         =  >         >  ?         ?  @         @  #B         $B  KC         LC  sD         tD  E         E  F          g  )h          Rh  ei          i  j          j  k          l  m         Bm  in         jn  o         o  p         p  q         q  	s         
s  1t         2t  Yu         Zu  v         v  w         w  x         x  y         y  !{         "{  I|         J|  q}         r}  ~         ~                                   9         :  a         b                        ه         ڇ             )         *  Q         R  y         z             Ɏ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               F!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     N
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-5-x    5-C-|    C-v-\    v--    --h    -ub-    ub-x-	    x--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               I!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        3  @          f  s                                              &  "          M  I	          t	  p
          
                                              )         *  C                      4         5  K         L  b         c  y                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         ,            J  <          h  Z            x                     "  $         $  *%         ,%  J&         L&  j'         l'  (         (  )         )  *         *  +         +  
-         -  *.         ,.  J/         L/  j0         l0  1         1  2         2  3         3  4         4  
6         6  *7         ,7  J8         L8  j9         l9  :         :  ;         ;  <         <  =         =  
?         ?  *@         ,@  JA         LA  jB         lB  C         C  D         D  E         AF  _G         aG  H         H  I         I  J         J  K         K  L         M  N         !N  ?O         AO  _P         aP  Q         Q  R         R  S         S  T         T  U         V  W         !W  ?X         AX  _Y         aY  Z         Z  [         [  \         \  ]         ]  ^         _  `         !`  ?a         Aa  _b         ab  c         c  d         d  e         e  f         f  g         h  i         i  k         k  1l         2l  Qm         Rm  qn         rn  o         o  p         p  q         q  r         r  t         t  1u         2u  Qv         Rv  qw         rw  x         x  y         y  z         z  {         {  }         }  1~         2~  Q         R  q         r                        у         ҃                        1         2  Q         R  q                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               =!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 y
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -;-    ;-:-x    :-I-|    I-|-\    |--    --h    -Kb-    Kb-!-	    !--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               J!         X!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      ,         -  C	         D	  Z
         [
  q         5  N         O  h         i                                                                      ,  .          S  U          z                                        !          F!  S"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       5         <  S         Z  q         x                                              
         
  %         ,  C         J  a         h                                                                    3         :  Q         X  o         v                                                         #         *  A         H  _          f   }!         p"  #         #  $         $  %         %  &         &  (         (  )         $)  =*         B*  [+         `+  y,         ~,  -         -  .         .  /         /  0         0  2         2  -3         23  K4         P4  i5         n5  6         6  7         7  8         8  9         9  :         ;  <         "<  ;=         @=  Y>         ^>  w?         |?  @         @  A         A  B         B  C         C  E         E  +F         0F  IG         :H  SI         XI  qJ         vJ  K         K  L         L  M         M  N         N  P         P  %Q         *Q  CR         HR  aS         fS  T         T  U         U  V         V  W         W  X         X  Z         Z  3[         8[  Q\         V\  o]         t]  ^         ^  _         _  `         `  a         a  c         
c  #d         (d  Ae         Fe  _f          j  k          k  l          m  n          6n  Bo          ho  tp                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               #!         6!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   K      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-5-x    5-C-|    C-v-\    v--    --h    -ub-    ub-x-	    x--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               E!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        3  A          f  t                                              &  !          M  H	          t	  o
          
                                              %         &  >                      /         0  F         G  ]         ^  t                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                                         :  ,         "  
$         $  +%         ,%  K&         L&  k'         l'  (         (  )         )  *         *  +         +  -         -  +.         ,.  K/         L/  k0         l0  1         1  2         2  3         3  4         4  6         6  +7         ,7  K8         L8  k9         l9  :         :  ;         ;  <         <  =         =  ?         ?  +@         ,@  KA         LA  kB         lB  C         C  D         D  E         AF  `G         aG  H         H  I         I  J         J  K         K   M         M   N         !N  @O         AO  `P         aP  Q         Q  R         R  S         S  T         T   V         V   W         !W  @X         AX  `Y         aY  Z         Z  [         [  \         \  ]         ]   _         _   `         !`  @a         Aa  `b         ab  c         c  d         d  e         e  f         f   h         h  i         i  k         k  0l         2l  Pm         Rm  pn         rn  o         o  p         p  q         q  r         r  t         t  0u         2u  Pv         Rv  pw         rw  x         x  y         y  z         z  {         {  }         }  0~         2~  P         R  p         r                        Ѓ         ҃                        0         2  P         R  p                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               =!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 y
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-:-x    :-{I-|    {I-|-\    |--    --h    -Qb-    Qb-!-	    !--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               H!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      ,         -  C	         D	  Z
         [
  q         5  M         N  f         g                                                                       '  (          N  O          u                                       !          A!  O"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       4         <  R         Z  p         x                                              
         
  $         ,  B         J  `         h  ~                                                                  2         :  P         X  n         v                                                         "         *  @         H  ^          f   |!         p"  #         #  $         $  %         %  &         &  (         (   )         $)  >*         B*  \+         `+  z,         ~,  -         -  .         .  /         /  0         0  2         2  .3         23  L4         P4  j5         n5  6         6  7         7  8         8  9         9   ;         ;  <         "<  <=         @=  Z>         ^>  x?         |?  @         @  A         A  B         B  C         C  E         E  ,F         0F  JG         :H  TI         XI  rJ         vJ  K         K  L         L  M         M  N         N  P         P  &Q         *Q  DR         HR  bS         fS  T         T  U         U  V         V  W         W  X         X  Z         Z  4[         8[  R\         V\  p]         t]  ^         ^  _         _  `         `  a         a  c         
c  $d         (d  Be         Fe  `f          p  q          q  r          r  
t          0t  <u          bu  nv                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               '!         :!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   L      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-m-     m--    -/;-x    /;-I-|    I-}-\    }--    --h    -ob-    ob-?-	    ?--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               7!         @!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         1  G         H  ^         _  u	         v	  
         
           5  ]         ^                                                                     <  7          Y  T          v  q                                    '           J   d!          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       :         <  X         Z  v         x                                              
         
  *         ,  H         J  f         h                                                                    8         :  V         X  t         v                                              
           (         *  F         H  d          f   !         p"  #         #  $         $  %         %  &         &  (         (  ")         $)  @*         B*  ^+         `+  |,         ~,  -         -  .         .  /         /  0         0  2         2  03         23  N4         P4  l5         n5  6         6  7         7  8         8  9         9  ;         ;   <         "<  >=         @=  \>         ^>  z?         |?  @         @  A         A  B         B  C         C  E         E  .F         0F  LG         :H  VI         XI  tJ         vJ  K         K  L         L  M         M  N         N  
P         P  (Q         *Q  FR         HR  dS         fS  T         T  U         U  V         V  W         W  X         X  Z         Z  6[         8[  T\         V\  r]         t]  ^         ^  _         _  `         `  a         a  c         
c  &d         (d  De         Fe  bf          /  O          r              Չ                      ;  [                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               6!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
  l      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     --    -<-x    <-J-|    J-~-\    ~--    --h    -3b-    3b--	    --h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               <!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        3         4  M         N  g         h                                      &          L  Y                                              ,  @          i  }                                  '          (   J!         K!  m"         n"  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       5         :  R         W  o         t                                               
         
           "  :         ?  W         \  t         y                                                       
  "         '  ?         D  \         a  y         ~                                              
           '         ,  D          I   a!         d#  ~$         $  %         %  &         &  '         '  )         )  '*         ,*  K+         P+  o,         t,  -         -  .         .  /         /  0         1  !2         &2  C3         H3  e4         j4  5         5  6         6  7         7  8         8  :         :  1;         6;  S<         X<  s=         x=  >         >  ?         ?  @         @  A         A  C         C  3D         8D  SE         XE  sF         xF  G          MV  jW          W  X          X  Y          Z  3[          Y[  v\         m  n         n  9o         ;o  Up         Wp  qq         sq  r         r  s         s  t         t  u         u  v         v  x         x  5y         7y  Qz         Sz  m{         o{  |         |  }         }  ~         ~                                   1         3  M         O  i         k                                   ى         ۉ                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               !         *!  1         1  v=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   K      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-s-     s--    -;;-x    ;;-I-|    I-#}-\    #}--    --h    -cb-    cb-?-	    ?--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               9!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         1  G         H  ^         _  u	         v	  
         
           :  c         d                                                         )  #          F  @          c  ]            z                                    2           T   o!          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       ;         <  Y         Z  w         x                                              
         
  +         ,  I         J  g         h                                                                    9         :  W         X  u         v                                                         )         *  G         H  e          f   !         p"  #         #  $         $  %         %  &         &  (         (  #)         $)  A*         B*  _+         `+  },         ~,  -         -  .         .  /         /  0         0  2         2  13         23  O4         P4  m5         n5  6         6  7         7  8         8  9         9  ;         ;  !<         "<  ?=         @=  ]>         ^>  {?         |?  @         @  A         A  B         B  C         C  E         E  /F         0F  MG         :H  WI         XI  uJ         vJ  K         K  L         L  M         M  N         N  P         P  )Q         *Q  GR         HR  eS         fS  T         T  U         U  V         V  W         W  X         X  Z         Z  7[         8[  U\         V\  s]         t]  ^         ^  _         _  `         `  a         a  	c         
c  'd         (d  Ee         Fe  cf          ~                          $          G  g                               >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               2!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
  m      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-y-     y--    -<-x    <-J-|    J-~-\    ~--    --h    -9b-    9b--	    --h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               :!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        1         2  J         K  c         d  |                                    "          G  U          z                                    '  <          d  y                                   $          %   H!         I!  l"         m"  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       6         :  S         W  p         t                                              
         
           "  ;         ?  X         \  u         y                                                       
  #         '  @         D  ]         a  z         ~                                                         (         ,  E          I   b!         d#  $         $  %         %  &         &  '         '  )         )  (*         ,*  L+         P+  p,         t,  -         -  .         .  /         /   1         1  "2         &2  D3         H3  f4         j4  5         5  6         6  7         7  8         8  :         :  2;         6;  T<         X<  t=         x=  >         >  ?         ?  @         @  A         A  C         C  4D         8D  TE         XE  tF         xF  G          \  ]          ]  ^          "_  ?`          e`  a          a  b         m  n         n  :o         ;o  Vp         Wp  rq         sq  r         r  s         s  t         t  u         u  v         v  x         x  6y         7y  Rz         Sz  n{         o{  |         |  }         }  ~         ~                                   2         3  N         O  j         k                                   ډ         ۉ                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                                !         .!  1         1  ~=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   L      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-7<-x    7<-J-|    J-~-\    ~--    --h    -Qb-    Qb-9-	    9--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               M!         V!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             (         )  Q         R  z         {                                   	  #          F  `                                    
          0  =          c  p                                 -  F         G  `          a   z!         {!  "         "  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       8         :  U         W  r         t                                              
         
            "  =         ?  Z         \  w         y                                                       
  %         '  B         D  _         a  |         ~                                                         *         ,  G          I   d!         d#  $         $  %         %  &         &  '         '  )         )  **         ,*  N+         P+  r,         t,  -         -  .         .  /         /  1         1  $2         &2  F3         H3  h4         j4  5         5  6         6  7         7  8         8  :         :  4;         6;  V<         X<  v=         x=  >         >  ?         ?  @         @  A         A  C         C  6D         8D  VE         XE  vF         xF  G          I  J          J  K          L  M          3M  ?N          _N  kO         m  n         n  :o         ;o  Vp         Wp  rq         sq  r         r  s         s  t         t  u         u  v         v  x         x  6y         7y  Rz         Sz  n{         o{  |         |  }         }  ~         ~                                   2         3  N         O  j         k                                   ډ         ۉ                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               -!         :!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
  l      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-m-     m--    -:-x    :-uI-|    uI-|-\    |-5-    5--P    -cb-    cb-c-	    c-g-D
    g--׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               3!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      ,         -  C	         D	  Z
         [
  q         5  N         O  h         i                                                                      ,  .          S  U          z                                        !          F!  S"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p                     "  =         D  _         f                                              		         	  +
         2
  M         T  o         v                                                          ;         B  ]         d                                                         )         0  K         R  m         t                                     !         !  #         #  2$         7$  Q%         V%  p&         u&  '         '  (         (  )         )  +         +  6,         =,  ^-         e-  .         .  /         /  0         0  1         2  (3         /3  R4         Y4  |5         5  6         6  7         7  8         9  $:         +:  N;         U;  x<         <  =         =  >         >  ?         @  &A         -A  RB         YB  ~C         C  D         D  E         E  G         	G  .H         5H  OI           d  e          @e  ]f          f  g          g  h           i  j          n  o         o  9p         >p  Xq         ]q  wr         |r  s         s  t         t  u         u  v         v  x         x  1y         6y  Pz         Uz  o{         t{  |         |  }         }  ~         ~             
           )         .  H         M  g         l                        ć                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               5!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 }
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-7<-x    7<-J-|    J-~-\    ~--    --h    -Kb-    Kb-9-	    9--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               K!         V!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             )         *  S         T  }         ~                                     )          K  f                                              5  C          h  v                                 2  J         K  c          d   |!         }!  "         "  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       9         :  V         W  s         t                                              
         
  !         "  >         ?  [         \  x         y                                              	         
  &         '  C         D  `         a  }         ~                                                         +         ,  H          I   e!         d#  ~$         $  %         %  &         &  '         '  )         )  +*         ,*  O+         P+  s,         t,  -         -  .         .  /         /  1         1  %2         &2  G3         H3  i4         j4  5         5  6         6  7         7  8         8  :         :  5;         6;  W<         X<  w=         x=  >         >  ?         ?  @         @  A         A  C         C  7D         8D  WE         XE  wF         xF  G          O  P          P  Q          Q  R          S  T          ;T  GU         m  n         n  9o         ;o  Up         Wp  qq         sq  r         r  s         s  t         t  u         u  v         v  x         x  5y         7y  Qz         Sz  m{         o{  |         |  }         }  ~         ~                                   1         3  M         O  i         k                                   ى         ۉ                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               +!         :!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
  m      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-g-     g--    -:-x    :-uI-|    uI-|-\    |--    --P    -cb-    cb-c-	    c-m-D
    m--׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               2!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      ,         -  C	         D	  Z
         [
  q         5  M         N  f         g                                                                       '  (          N  O          u                                       !          A!  O"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p                     "  <         D  ^         f                                              	         	  *
         2
  L         T  n         v                                                          :         B  \         d  ~                                                       (         0  J         R  l         t                                     !         !  #         #  3$         7$  R%         V%  q&         u&  '         '  (         (  )         )  +         +  5,         =,  ]-         e-  .         .  /         /  0         0  1         2  '3         /3  Q4         Y4  {5         5  6         6  7         7  8         9  #:         +:  M;         U;  w<         <  =         =  >         >  ?         @  %A         -A  QB         YB  }C         C  D         D  E         E  G         	G  -H         5H  PI          @j  ]k          k  l          l  m          n  o         o  :p         >p  Yq         ]q  xr         |r  s         s  t         t  u         u  v         v  x         x  2y         6y  Qz         Uz  p{         t{  |         |  }         }  ~         ~                        *         .  I         M  h         l                        Ň             =          `  }                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               5!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 }
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -/-    /-);-    );-I-    I-}-d    }-/-    /--0    --    -b-H	    b-c-P
    c--
    -'׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               J!         X!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         1  G         H  ^         _  u	         v	  
         
           5  ]         ^                                                                     <  7          Y  T          v  q                                    '           J   d!          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q   w  operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       \                      "  B         D  d         f                                              	         	  0
         2
  R         T  t         v                                                          @         B  b         d                                                         .         0  P         R  r         t                                     !         !  #         #  5$         7$  T%         V%  s&         u&  '         '  (         (  )         )  +         +  ;,         =,  c-         e-  .         .  /         /  0         0  2         2  -3         /3  W4         Y4  5         5  6         6  7         7  8         9  ):         +:  S;         U;  }<         <  =         =  >         >  ?         @  +A         -A  WB         YB  C         C  D         D  E         E  G         	G  3H         5H  RI          n  o         o  <p         >p  [q         ]q  zr          |r  xs          s  t          t  u          u  v          v  w         x  4y         6y  Sz         Uz  r{         t{  |         |  }         }  ~         ~                        ,         .  K         M  j         l                        Ǉ                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               J!         X!  1         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,-[-     [--    -<-    <-J-    J-~-l    ~-	-    	--P    --    -b-h	    b--p
    --
    -հ-    հ-'׳-L                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               .!         >!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        3         4  M         N  g         h                                      &          L  Y                                              ,  @          i  }                                  '          (   J!         K!  m"         n"  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h         #  $          $  %         %  &         '  #(         *(  G)         N)  k*         r*  +         +  ,         ,  -         -  .         /  0         &0  C1         J1  g2         n2  3         3  4         4  5         5  6         6  8         "8  ?9         F9  c:         j:  ;         ;  <         <  =         =  >         >  @         @  ;A         BA  _B         fB  C         C  D         D  E         E  F         F  H          H  I          ;I  6J          \J  WK         }K  L         L  M         M  N         N  P         
P  ,Q         .Q  PR         RR  tS         vS  T         T  U         U  V         V  X         X  &Y         (Y  HZ         JZ  j[         l[  \         \  ]         ]  ^         ^  _         _  a         a  6b         8b  Xc         Zc  xd         zd  e         e  f         f  g         g  h         h  j         j  8k         :k  Xl         Zl  xm         zm  n         n  o         o  p         p  q         q  s         s  :t         ?t  [u         `u  |v         v  w         w  x         x  y         y   {         {  !|         &|  B}         G}  c~         h~                        Ɓ         ˁ                        )         .  J         O  k         p                                           b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               L!         V!  1         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    ^
         _
             (                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     C
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -/-    /-5;-    5;-I-    I-}-d    }-5-    5--0    --    -b-H	    b-c-P
    c--
    -'׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               I!         V!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         1  G         H  ^         _  u	         v	  
         
           :  c         d                                                         )  #          F  @          c  ]            z                                    2           T   o!          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q   w  operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       \            !         "  C         D  e         f                                              	         	  1
         2
  S         T  u         v                                                          A         B  c         d                                                         /         0  Q         R  s         t                                     !         !  #         #  6$         7$  U%         V%  t&         u&  '         '  (         (  )         )  +         +  <,         =,  d-         e-  .         .  /         /  0         0  2         2  .3         /3  X4         Y4  5         5  6         6  7         7   9         9  *:         +:  T;         U;  ~<         <  =         =  >         >   @         @  ,A         -A  XB         YB  C         C  D         D  E         E  G         	G  4H         5H  SI          n  o         o  =p         >p  \q         ]q  {r         |r  s         s  t         t  u         u  v         v  x          x  y          6y  2z          Uz  Q{          t{  p|          |  }         }  ~         ~                        -         .  L         M  k         l                        ȇ                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               H!         T!  1         1  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,-[-     [--    -<-    <-J-    J-~-l    ~-	-    	--P    --    -b-h	    b--p
    --
    -۰-    ۰-'׳-L                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               .!         >!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        1         2  J         K  c         d  |                                    "          G  U          z                                    '  <          d  y                                   $          %   H!         I!  l"         m"  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        #  $         $  %         %  &         '  "(         *(  F)         N)  j*         r*  +         +  ,         ,  -         -  .         /  0         &0  B1         J1  f2         n2  3         3  4         4  5         5  6         6  8         "8  >9         F9  b:         j:  ;         ;  <         <  =         =  >         >  @         @  :A         BA  ^B         fB  C         C  D         D  E         E  F         F  H         H  7I         ;I  XJ         \J  yK          }K  xL         L  M         M  N         N  	P         
P  -Q         .Q  QR         RR  uS         vS  T         T  U         U  V         V  X         X  'Y         (Y  IZ         JZ  k[         l[  \         \  ]         ]  ^         ^  _         _  a         a  7b         8b  Yc         Zc  yd         zd  e         e  f         f  g         g  h         h  j         j  9k         :k  Yl         Zl  ym         zm  n          n  o          o  p          p  q          q  r         s  ;t         ?t  \u         `u  }v         v  w         w  x         x  y         y  {         {  "|         &|  C}         G}  d~         h~                        ǁ         ˁ             	           *         .  K         O  l         p                                           b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               J!         V!  1         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    ^
         _
             )                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     D
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -5-    5-1<-    1<-J-    J-~-d    ~-	-    	--p    -b-    b-낳-
    낳--
    -ᰳ-t    ᰳ-!׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               L!         `!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             (         )  Q         R  z         {                                   	  #          F  `                                    
          0  =          c  p                                 -  F         G  `          a   z!         {!  "         "  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         H  O          r  y                                             #  $         $  %         %  '         '  ((         *(  L)         N)  p*         r*  +         +  ,         ,  -         -   /         /  $0         &0  H1         J1  l2         n2  3         3  4         4  5         5  6         6   8         "8  D9         F9  h:         j:  ;         ;  <         <  =         =  >         >  @         @  @A         BA  dB         fB  C         C  D         D  E         E  F         F  H         H  9I         ;I  ZJ         \J  {K         }K  L         L  M         M  N         N  	P         
P  -Q         .Q  QR         RR  uS         vS  T         T  U         U  V         V  X         X  'Y         (Y  IZ         JZ  k[         l[  \         \  ]         ]  ^         ^  _         _  a         a  7b         8b  Yc         Zc  yd         zd  e         e  f         f  g         g  h         h  j         j  9k         :k  Yl         Zl  ym         zm  n         n  o         o  p         p  q         q  s         s  =t         ?t  ^u         `u  v         v  w         w  x         x  y         y  {         {  $|         &|  E}         G}  f~         h~                        Ɂ         ˁ                        ,         .  M         O  n         p                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               e!         j!  2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    ^
         _
             &      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     P
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                 w        Sܯ--0     --     -c-     c-	B-P    	B-c-    c--    -!׳-p                             b      *      7   I      operation.0     sync             :            2            b         reason  ExchangeBarrier value   1       invocations     1                                     b      *      7   I      operation.0     sync             :            2            b         reason  ExchangeBarrier value   3       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   5       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   2       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -/-    /-1<-    1<-J-    J-~-d    ~-	-    	--p    -b-    b-낳-
    낳--
    -ᰳ-t    ᰳ-!׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               J!         \!  1         
2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             )         *  S         T  }         ~                                     )          K  f                                              5  C          h  v                                 2  J         K  c          d   |!         }!  "         "  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |           !          D  K          n  u              !          !  "         #  $         $  %         %  '         '  )(         *(  M)         N)  q*         r*  +         +  ,         ,  -         -  /         /  %0         &0  I1         J1  m2         n2  3         3  4         4  5         5  6         6  !8         "8  E9         F9  i:         j:  ;         ;  <         <  =         =  >         >  @         @  AA         BA  eB         fB  C         C  D         D  E         E  F         F  H         H  :I         ;I  [J         \J  |K         }K  L         L  M         M  N         N  P         
P  ,Q         .Q  PR         RR  tS         vS  T         T  U         U  V         V  X         X  &Y         (Y  HZ         JZ  j[         l[  \         \  ]         ]  ^         ^  _         _  a         a  6b         8b  Xc         Zc  xd         zd  e         e  f         f  g         g  h         h  j         j  8k         :k  Xl         Zl  xm         zm  n         n  o         o  p         p  q         q  s         s  >t         ?t  _u         `u  v         v  w         w  x         x  y         y  {         {  %|         &|  F}         G}  g~         h~                        ʁ         ˁ                        -         .  N         O  o         p                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               e!         j!  2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    ^
         _
             &      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     Q
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                 w        Sܯ--0     --     -c-     c-	B-P    	B-c-    c--    -!׳-p                             b      *      7   I      operation.0     sync             :            2            b         reason  ExchangeBarrier value   1       invocations     1                                     b      *      7   I      operation.0     sync             :            2            b         reason  ExchangeBarrier value   3       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   5       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   2       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -{,-h     {,-1-     1--    -5-    5-C-    C-v-l    v-30-    30--x    -]b-	    ]b-傳-
    傳--
    -U-|    U-ճ-    ճ-׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     h            	!          B!  1          1  {=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P              	          1  :          b  k                                                "          G  K	          p	  t
          
                                              %         &  >           $         %  =         >  V         W  o         p                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       6  G          l  }                                 T$  v%         x%  &         vL  M         M  N         N  O         O  Q         Q  )R         *R  MS         NS  qT         rT  U         U  V         V  W         W  Y         Y  %Z         &Z  I[         J[  m\         n\  ]         ]  ^         ^  _         _  `         `  !b         "b  Ec         Fc  id         jd  e         e  f         f  g         g  h         h  j         j  Ak         Bk  el         fl  m         m  n         n  o         o  p         p  r         r  =s         >s  at         bt  u         u  v         v  w         w  x         x  z         z  9{         :{  ]|         ^|  }         }  ~         ~                                   5         6  Y         Z  }         ~             Ň         Ƈ                        1         2  U         V  y         z                                            
  ,         .  P         R  t         v                                              (         *  L         N  p         r                        ܠ         ޠ              $         &  H         J  l         n                        ب         ڨ                       "  D         F  h         j                        ԰                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               C!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    `
         a
             )      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    `
         a
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,-I-     I--    -7<-    7<-J-    J-~-d    ~-[-    [--p    -b-    b--
    --
    -ɰ-t    ɰ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     u            !          J!  1          1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        1         2  J         K  c         d  |                                    #          K  T          |                                    +  G          j                         	         
  6          7   c!         d!  "         "  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       '  )          N  P          u  w                     $  %         %  &         &  (         (  ')         ()  M*         N*  s+         t+  ,         ,  -         -  .         .  0         0  11         21  W2         X2  }3         ~3  4         4  5         5  6         6  8         8  ;9         <9  a:         b:  ;         ;  <         <  =         =  >         >  @          @  EA         FA  kB         lB  C         C  D         D  E         E  G         G  (H         *H  NI         PI  tJ         vJ  K         K  L         L  M         M  O         O  2P         4P  XQ         ZQ  ~R         R  S         S  T         T  U         U  W         W  <X         >X  bY         dY  Z         Z  [         [  \         \  ]         ]   _         "_  F`         H`  la         na  b         b  c         c  d         d  f         
f  0g         2g  Xh         Zh  i         i  j         j  k         k  l         l   n         "n  Ho         Jo  np         pp  q         q  r         r  s         s  u         u  ,v         .v  Rw         Tw  xx         zx  y         y  z         z  {         {  }         }  6~         8~  \         ^                        ΂         Ђ                        @         B  f         h                        ؊         ڊ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               a!         b!  2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    `
         a
             6      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     ?
                   
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            8     -,-h     ,-7-     7--    -5-    5-C-    C-v-l    v-30-    30--x    -]b-	    ]b-낳-
    낳--
    -U-|    U-ճ-    ճ-׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     i            !          F!  1          1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P              
          1  ;          b  l                                                !          G  J	          p	  s
          
                                              %         &  >           $         %  =         >  V         W  o         p                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                     D  U          z  	          	  
          
           T$  w%         x%  &         vL  M         M  N         N  O         O  Q         Q  (R         *R  LS         NS  pT         rT  U         U  V         V  W         W   Y         Y  $Z         &Z  H[         J[  l\         n\  ]         ]  ^         ^  _         _  `         `   b         "b  Dc         Fc  hd         jd  e         e  f         f  g         g  h         h  j         j  @k         Bk  dl         fl  m         m  n         n  o         o  p         p  r         r  <s         >s  `t         bt  u         u  v         v  w         w  x         x  z         z  8{         :{  \|         ^|  }         }  ~         ~                                   4         6  X         Z  |         ~             ć         Ƈ                        0         2  T         V  x         z                                   	         
  -         .  Q         R  u         v                                              )         *  M         N  q         r                        ݠ         ޠ             %         &  I         J  m         n                        ٨         ڨ             !         "  E         F  i         j                        հ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               C!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    `
         a
             *      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    `
         a
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,-I-     I--    -=<-    =<-J-    J-%~-d    %~-a-    a--p    -b-    b--
    --
    -ɰ-t    ɰ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     t            !          J!  1          1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        1         2  J         K  c         d  |                                    $          K  U          |                          	          +  H          j                         
           8          9   f!         g!  "         "  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                   	          8	  :
          _
  a         $  %         %  &         &   (         (  &)         ()  L*         N*  r+         t+  ,         ,  -         -  .         .  
0         0  01         21  V2         X2  |3         ~3  4         4  5         5  6         6  8         8  :9         <9  `:         b:  ;         ;  <         <  =         =  >         >  @          @  DA         FA  jB         lB  C         C  D         D  E         E  G         G  )H         *H  OI         PI  uJ         vJ  K         K  L         L  M         M  O         O  3P         4P  YQ         ZQ  R         R  S         S  T         T  U         U  W         W  =X         >X  cY         dY  Z         Z  [         [  \         \  ]         ]  !_         "_  G`         H`  ma         na  b         b  c         c  d         d  	f         
f  1g         2g  Yh         Zh  i         i  j         j  k         k  l         l  !n         "n  Io         Jo  op         pp  q         q  r         r  s         s  u         u  -v         .v  Sw         Tw  yx         zx  y         y  z         z  {         {  }         }  7~         8~  ]         ^                        ς         Ђ                        A         B  g         h                        ي         ڊ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               a!         b!  2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    `
         a
             7      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     @
                   
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,-7-     7--    -<-x    <-J-|    J-}-\    }-]3-    ]3--h    -ob-    ob-m-	    m--h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     h            
!          D!  1          1  ~=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             "         #  E         F  h         i                                               "  4          a  s                                              A  J          r  {                       '         (  @          A   Y!         Z!  r"         s"  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                      '  G         N  n         u                                   
           1	         8	  X
         _
                                                       "  B         I  i         p                                              ,         3  S         Z  z                                                       =          D   d!         k!  "          x2  3          3  4          4  6          A6  \7          7  8         K  L         L  N         N  5O         >O  cP         lP  Q         Q  R         R  S         S  U         $U  IV         RV  wW         W  X         X  Y         Y  Z         [  '\         0\  S]         \]  ^         ^  _         _  `         `  b         b  /c         8c  Yd         bd  e         e  f         f  g         g  i         
i  +j         4j  Uk         ^k  l         l  m         m  n         ȑ                        0         4  T         X  x         |                      ę                        ,         0  P         T  t         x                                              (         ,  L         P  p         t                        ܪ                       $         (  H         L  l         p                        ز         ܲ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               F!         J!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     <
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            8     -,-h     ,-s-     s--    -/5-    /5-C-    C-w-l    w---    ---x    -Kb-	    Kb--
    --
    -g-|    g-ճ-    ճ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               =!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        ?  [          ~                                    ;  6          V  Q          q  l	          	  
          
                                   H         I  u         v           4  L         M  e         f  ~                                                    b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |        V"  {#         }#  $          0  1          1  2          2  4          .4  @5          `5  r6         H  I         I  J         J  K         K  M         M  EN         GN  mO         oO  P         P  Q         Q  R         R  T         T  5U         7U  ]V         _V  W         W  X         X  Y         Y  Z         Z  %\         '\  M]         O]  u^         w^  _         _  `         `  a         a  c         c  =d         ?d  ee         ge  f         f  g         g  h         h  j         j  -k         /k  Ul         Wl  }m         m  n         n  o         o  p         p  r         r  Es         Gs  mt         ot  u         u  v         v  w         w  y         y  5z         7z  ]{         _{  |         |  }         }  ~         ~             %         '  M         O  u         w             Ņ         ǅ                        =         ?  e         g                        ݍ         ߍ             ,         .  S         U  {         }             ˔         ͔                        C         E  k         m                                              3         5  [         ]                        ӣ         գ             #         %  K         M  s         u             ê         Ū                        ;                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     u            !          P!  1          1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    `
         a
             6      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    b
         c
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            0     -,-X     ,-7-     7--    -<-x    <-J-|    J-}-\    }-c3-    c3--h    -ub-    ub-m-	    m--h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     i            !          D!  1          1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             #         $  G         H  k         l                                               '  :          f  y                                              F  P          w                         ,         -  E          F   ^!         _!  w"         x"  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            !         '  H         N  o         u                                              2	         8	  Y
         _
                                                       "  C         I  j         p                                              -         3  T         Z  {                                                       >          D   e!         k!  "          8  9          
:  %;          M;  h<          <  =          =  >         K  L         L  N         N  4O         >O  bP         lP  Q         Q  R         R  S         S  U         $U  HV         RV  vW         W  X         X  Y         Y  Z         [  &\         0\  R]         \]  ~^         ^  _         _  `         `  b         b  .c         8c  Xd         bd  e         e  f         f  g         g   i         
i  *j         4j  Tk         ^k  ~l         l  m         m  n         ȑ                        1         4  U         X  y         |                      ę             	           -         0  Q         T  u         x                                              )         ,  M         P  q         t                        ݪ                      %         (  I         L  m         p                        ٲ         ܲ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               H!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     ;
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            8     -,-h     ,-s-     s--    -;5-    ;5-C-    C-#w-l    #w---    ---x    -Eb-	    Eb--
    --
    -m-|    m-ճ-    ճ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ;!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        ?  \          ~                                    ;  5          V  P          q  k	          	  
          
                                   K         L  y         z           >  V         W  o         p                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |        V"  |#         }#  $          6  7          7  8          8  :          (:  :;          Z;  l<         H  I         I  J         J  K         K  M         M  FN         GN  nO         oO  P         P  Q         Q  R         R  T         T  6U         7U  ^V         _V  W         W  X         X  Y         Y  Z         Z  &\         '\  N]         O]  v^         w^  _         _  `         `  a         a  c         c  >d         ?d  fe         ge  f         f  g         g  h         h  j         j  .k         /k  Vl         Wl  ~m         m  n         n  o         o  p         p  r         r  Fs         Gs  nt         ot  u         u  v         v  w         w  y         y  6z         7z  ^{         _{  |         |  }         }  ~         ~             &         '  N         O  v         w             ƅ         ǅ                        >         ?  f         g                        ލ         ߍ             -         .  T         U  |         }             ̔         ͔                        D         E  l         m                                              4         5  \         ]                        ԣ         գ             $         %  L         M  t         u             Ī         Ū                        <                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     t            !          L!  1          1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    `
         a
             7      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    c
         d
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            0     -,-X     ,-a-     a--    -5-x    5-C-|    C-v-\    v--    --h    -9b-    9b-n-	    n--h
    -!׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               /!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P              	          1  :          b  k                                                "          G  K	          p	  t
          
                                              %         &  >           $         %  =         >  V         W  o         p                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                     %  C         J  h         o                                              !	         (	  F
         M
  k         r                                              $         +  I         P  n         u                                            	  '         .  L         S  q         x                                                *!         1!  O"          <  =          @=  T>          |>  ?          ?  @          @  B         I  J         J  K         K  L         M  !N         (N  FO         MO  kP         rP  Q         Q  R         R  S         S  T         U  $V         +V  IW         PW  nX         uX  Y         Y  Z         Z  [         [  ]         	]  '^         .^  L_         S_  q`         x`  a         a  b         b  c         c  	e         e  /f         6f  Ug         \g  {h         h  i         i  j         j  k         k  m         m  ;n         Dn  co         lo  p         p  q         q  r         r  t         t  +u         4u  Sv         \v  {w         w  x         x  y         y  z         z  |         $|  C}         L}  k~         t~                      Ā                        3         <  [         d                        ӈ         ܈             #         ,  K         T  s         |                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     l            !          H!  1          1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     <
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-C<-x    C<-J-|    J-+~-\    +~-.-    .--h    -]b-    ]b-n-	    n--h
    -!׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               B!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        1         2  J         K  c         d  |                                    #          K  T          |                                    +  G          j                         	         
  6          7   c!         d!  "         "  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            (         *  R         T  |         ~                                   $         &  N	         P	  x
         z
                                             "  J         L  t         v                                              F         H  p         r                                              B         D  l         n               !          M;  m<          <  =          =  >          ?  6@          Y@  yA         @G  iH         jH  I         I  J         J  K         K  "M         $M  LN         NN  vO         xO  P         P  Q         Q  R         R  T          T  HU         JU  rV         tV  W         W  X         X  Y         Y  [         [  D\         F\  n]         p]  ^         ^  _         _  `         `  b         b  @c         Bc  jd         ld  e         e  f         f  g         g  i         i  <j         >j  fk         hk  l         l  m           ͏         Ώ             !         "  K         L  u         v             ɖ         ʖ                        G         H  q         r             ŝ         Ɲ                        C         D  m         n                      ¤                        ?         @  i         j                                              ;                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     w            $!          Z!  1          2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     O
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-a-     a--    -5-x    5-C-|    C-v-\    v--    --h    -3b-    3b-n-	    n--h
    -!׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               1!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P              
          1  ;          b  l                                                !          G  J	          p	  s
          
                                              %         &  >           $         %  =         >  V         W  o         p                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                     %  D         J  i         o                                              "	         (	  G
         M
  l         r                                               %         +  J         P  o         u                                            	  (         .  M         S  r         x                                                +!         1!  P"          0B  DC          lC  D          D  E          E  F           G  4H         I  J         J  K         K  L         M  "N         (N  GO         MO  lP         rP  Q         Q  R         R  S         S   U         U  %V         +V  JW         PW  oX         uX  Y         Y  Z         Z  [         [  ]         	]  (^         .^  M_         S_  r`         x`  a         a  b         b  c         c  
e         e  0f         6f  Vg         \g  |h         h  i         i  j         j  k         k  m         m  :n         Dn  bo         lo  p         p  q         q  r         r  t         t  *u         4u  Rv         \v  zw         w  x         x  y         y  z         z  |         $|  B}         L}  j~         t~                      Ā             
           2         <  Z         d                        ҈         ܈             "         ,  J         T  r         |                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     k            !          D!  1          1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     ;
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -W -X     W -հ-     հ-ܰ-    ܰ-5-p    5-@-T    @-Ci-4    Ci-.-    .--8    -]b-    ]b-n-	    n--8
    -!׳-
                   =         b      J      7   U   7   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               B!         T!  P1                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 64 })  outputElements  5184    input0Elements  31104   input1Elements  24576   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     4                           #         b      J      7   M     operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       @                        1         2  J         K  c          `	  j
          
                                    $  A          c                                                ?         @  m         n                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     4                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  1728    input0Elements  1728    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       x            )         *  S         T  }         ~                                   %         &  O	         P	  y
         z
                                   !         "  K         L  u         v                                              G         H  q         r                                              C         D  m         n               !          A  B          B  C         @G  hH         jH  I         I  J         J  K         K  #M         $M  MN         NN  wO         xO  P         P  Q         Q  R         R  T          T  IU         JU  sV         tV  W         W  X         X  Y         Y  [         [  E\         F\  o]         p]  ^         ^  _         _  `         `  b         b  Ac         Bc  kd         ld  e         e  f         f  g         g  i         i  =j         >j  gk         hk  l         l  m          m  n           o   p           ̏         Ώ                       "  J         L  t         v             Ȗ         ʖ                        F         H  p         r             ĝ         Ɲ                        B         D  l         n                      ¤                        >         @  h         j                                              :                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     v            #!          Z!  1          2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     N
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-s-     s--    -G;-x    G;-I-|    I-/}-\    /}--    --h    --b-    -b-x-	    x--h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               7!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         ;  S         T  l         m  	         	  
         
           {                                              )          *            E  ,          `  G          {  b            }                                  /  A           n   !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                        <         @  \         `  |                                                      	  
          
  <         @  \         `  |                                                                    <         @  \         `  |                                                                    <         @  \         `  |                        !         !  "         "  $         $  $%         +%  G&         N&  j'         q'  (         (  )         )  *         *  +         +  -          -  <.         C.  _/         f/  0         0  1         1  2         2  3         3  5         5  16         86  T7         [7  w8         ~8  9         9  :         :  ;         ;  =         
=  &>         ->  I?         P?  l@         s@  A         A  B         B  C         "D  >E         EE  aF         hF  G         G  H          H  I         I  J         K  L         &L  BM         IM  eN         lN  O         O  P         P  Q         Q  R         R  T         T  7U         >U  ZV         aV  }W         W  X         X  Y         Y  Z         Z  	\         \  ,]         3]  O^         V^  r_         y_  `         `  a         a  b         b  c         d  !e         (e  Df         Kf  gg          l  m          m  n          n  o           p  %q                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     h            	!          B!  1          1  {=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 w
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-55-x    55-C-|    C-w-\    w-
-    
--h    -b-    b-x-	    x--h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               D!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        ?  [          ~                                    ;  6          V  Q          q  l	          	  
          
                                   H         I  u         v           4  L         M  e         f  ~                                                    b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       .  9          \  g                                 e$  %         %  &         &  '         '  (         (  *         
*  )+         ++  J,         L,  k-         m-  .         .  /         /  0         0  1         1  3         3  14         34  R5         T5  s6         u6  7         7  8         8  9         9  :         :  <         <  9=         ;=  Z>         \>  {?         }?  @         @  A         A  B         B  C         D   E         "E  AF         F  G         G  I         
I  )J         +J  JK         LK  kL         mL  M         M  N         N  O         O  P         P  R         R  1S         3S  RT         TT  sU         uU  V         V  W         W  X         X  Y         Y  [         [  9\         ;\  Z]         \]  {^         }^  _         _  `         `  a         a  b         c   d         "d  Ae         Ce  bf         df  g         g  h         h   j         "j  Dk         Fk  hl         jl  m         m  n         n  o         o  p         p  r         r  @s         Bs  dt         ft  u         u  v         v  w         w  x         x  z         z  <{         >{  `|         b|  }         }  ~         ~                                   8         :  \         ^                        ȇ         ʇ                        4                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               H!         X!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-m-     m--    -S;-x    S;-I-|    I-;}-\    ;}--    --h    -9b-    9b-x-	    x--h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               5!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         ;  S         T  l         m  	         	  
         
                                                         3          4            O  5          j  P            k                                              9  L           x   !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                        =         @  ]         `  }                                                      	  
          
  =         @  ]         `  }                                                                    =         @  ]         `  }                                                                    =         @  ]         `  }                        !         !  "         "  $         $  %%         +%  H&         N&  k'         q'  (         (  )         )  *         *  +         +  -          -  =.         C.  `/         f/  0         0  1         1  2         2  3         3  5         5  26         86  U7         [7  x8         ~8  9         9  :         :  ;         ;  =         
=  '>         ->  J?         P?  m@         s@  A         A  B         B  C         "D  ?E         EE  bF         hF  G         G  H         I  J         K   L         &L  CM         IM  fN         lN  O         O  P         P  Q         Q  R         R  T         T  8U         >U  [V         aV  ~W         W  X         X  Y         Y  Z         Z  
\         \  -]         3]  P^         V^  s_         y_  `         `  a         a  b         b  c         d  "e         (e  Ef         Kf  hg          Rq  Wr          r  s          s  t          t  u          v  w                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     i            !          F!  1          1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 w
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-A5-x    A5-C-    C-)w-
     )w-
-z     
--R
     -{b-     {b-x-     x--     -'׳-r                    =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               D!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7         operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1                               b                    operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   R     A   z   E   "   
8    r                                  reason  OperatorKernel  value   3       kernelSpec      outputElements  432        G  P                        ?  \          ~                                    ;  5          V  P          q  k	          	  
          
                                   K         L  y         z           >  V         W  o         p                                 "  Planned(ReductionSum { inputs: 3 })        z   input1Elements     "   432        z   input2Elements     "   432        b   invocations           5                                                  b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                                 3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                 B  M	          p	  {
          
           e$  %         %  &         &  '         '  (         (  	*         
*  *+         ++  K,         L,  l-         m-  .         .  /         /  0         0  1         1  3         3  24         34  S5         T5  t6         u6  7         7  8         8  9         9  :         :  <         <  :=         ;=  [>         \>  |?         }?  @         @  A         A  B         B   D         D  !E         "E  BF         F  G         G  	I         
I  *J         +J  KK         LK  lL         mL  M         M  N         N  O         O  P         P  R         R  2S         3S  ST         TT  tU         uU  V         V  W         W  X         X  Y         Y  [         [  :\         ;\  []         \]  |^         }^  _         _  `         `  a         a   c         c  !d         "d  Be         Ce  cf         df  g         g  h         h  !j         "j  Ek         Fk  il         jl  m         m  n         n  o         o  p         p  r         r  As         Bs  et         ft  u         u  v         v  w         w  x         x  z         z  ={         >{  a|         b|  }         }  ~         ~                                   9         :  ]         ^                        ɇ         ʇ                        5                         >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               D!         T!  1         1  =                                    b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1                    	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
              
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                           b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-m-     m--    -5-x    5-C-|    C-v-\    v-
-    
--h    -?b-    ?b-x-	    x--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               1!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P              	          1  :          b  k                                                "          G  K	          p	  t
          
                                              %         &  >           $         %  =         >  V         W  o         p                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                             <  B          h  n         e$  %         %  &         &  '         '  (         (  *         
*  $+         ++  E,         L,  f-         m-  .         .  /         /  0         0  1         1  3         3  ,4         34  M5         T5  n6         u6  7         7  8         8  9         9  :         :  <         <  4=         ;=  U>         \>  v?         }?  @         @  A         A  B         B  C         D  E         "E  <F         F  G         G  I         
I  $J         +J  EK         LK  fL         mL  M         M  N         N  O         O  P         P  R         R  ,S         3S  MT         TT  nU         uU  V         V  W         W  X         X  Y         Y  [         [  4\         ;\  U]         \]  v^         }^  _         _  `         `  a         a  b         c  d         "d  <e         Ce  ]f         df  ~g         g  h         h  j         "j  =k         Fk  al         jl  m         m  n         n  o         o  p         p  r         r  9s         Bs  ]t         ft  u         u  v         v  w         w  x         x  z         z  5{         >{  Y|         b|  }}         }  ~         ~                                   1         :  U         ^  y                               ʇ             	           -                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               %!         4!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 w
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -;-    ;-:-x    :-{I-|    {I-|-\    |-+-    +--h    -Eb-    Eb--	    --h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               L!         \!  1         
2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      &         '  ?	         @	  X
         Y
  q         ?  W         X  p         q                                                                      7  1          `  Z                                                 %!          M!  V"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       7         8  S         T  o         p                                              	         	             3         4  O         P  k         l                                                                    /         0  K         L  g         h                                                                    +          ,   G!         !  "         "  #         #  %         %  :&         <&  Y'         ['  x(         z(  )         )  *         *  +         +  ,         ,  .         .  2/         4/  Q0         S0  p1         r1  2         2  3         3  4         4  5         5  7         7  *8         ,8  I9         K9  h:         j:  ;         ;  <         <  =         =  >         >  @         @  "A         $A  AB         CB  `C         C  E         E  4F         6F  SG         UG  rH         tH  I          x  y          y  z          z  {          |   }          H}  N~         (  E         G  d         f                                 Õ                                    =         ?  \         ^  {         }                        ؟         ڟ                        5         7  T         V  s         u                        Ш         Ҩ                        -                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               '!         8!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   C      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-g-     g--    -5-x    5-C-|    C-v-\    v-
-    
--h    -Kb-    Kb-x-	    x--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               /!         @!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P              
          1  ;          b  l                                                !          G  J	          p	  s
          
                                              %         &  >           $         %  =         >  V         W  o         p                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                                         D  J         e$  %         %  &         &  '         '  (         (  *         
*  %+         ++  F,         L,  g-         m-  .         .  /         /  0         0  1         1  3         3  -4         34  N5         T5  o6         u6  7         7  8         8  9         9  :         :  <         <  5=         ;=  V>         \>  w?         }?  @         @  A         A  B         B  C         D  E         "E  =F         F  G         G  I         
I  %J         +J  FK         LK  gL         mL  M         M  N         N  O         O  P         P  R         R  -S         3S  NT         TT  oU         uU  V         V  W         W  X         X  Y         Y  [         [  5\         ;\  V]         \]  w^         }^  _         _  `         `  a         a  b         c  d         "d  =e         Ce  ^f         df  g         g  h         h  j         "j  <k         Fk  `l         jl  m         m  n         n  o         o  p         p  r         r  8s         Bs  \t         ft  u         u  v         v  w         w  x         x  z         z  4{         >{  X|         b|  |}         }  ~         ~                                   0         :  T         ^  x                               ʇ                        ,                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               )!         8!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 w
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-:-x    :-{I-|    {I-|-\    |-+-    +--h    -Kb-    Kb--	    --h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               H!         X!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      &         '  ?	         @	  X
         Y
  q         ?  W         X  p         q                                                                      7  0          `  Y                                                 &!          M!  W"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       6         8  R         T  n         p                                              	         	             2         4  N         P  j         l                                                                    .         0  J         L  f         h                                                                    *          ,   F!         !  "         "  #         #  %         %  ;&         <&  Z'         ['  y(         z(  )         )  *         *  +         +  ,         ,  .         .  3/         4/  R0         S0  q1         r1  2         2  3         3  4         4  5         5  7         7  +8         ,8  J9         K9  i:         j:  ;         ;  <         <  =         =  >         >  @         @  #A         $A  BB         CB  aC         C  E         E  5F         6F  TG         UG  sH         tH  I          v~  |                      Ҁ  ؁                       .  4         (  F         G  e         f                                 Õ                                     >         ?  ]         ^  |         }                        ٟ         ڟ                        6         7  U         V  t         u                        Ѩ         Ҩ                        .                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               )!         <!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   D      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-[-     [--    -M;-x    M;-I-|    I-5}-\    5}-+-    +--h    -Wb-    Wb-9-	    9--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               (!         8!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         ;  S         T  l         m  	         	  
         
           {                                              )          *            E  ,          `  G          {  b            }                                  /  A           n   !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       4         8  P         T  l         p                                              	         	             0         4  L         P  h         l                                                                    ,         0  H         L  d         h                                                                    (          ,   D!         !  "         "  #         #  %         %  5&         <&  T'         ['  s(         z(  )         )  *         *  +         +  ,         ,  .         .  -/         4/  L0         S0  k1         r1  2         2  3         3  4         4  5         5  7         7  %8         ,8  D9         K9  c:         j:  ;         ;  <         <  =         =  >         >  ?         @  A         $A  <B         CB  [C         C  E         E  /F         6F  NG         UG  mH         tH  I          b  c          c  d          e  )f          Sf  ig          g  h         (  @         G  _         f  ~                               Õ  ۖ                                  8         ?  W         ^  v         }                        ӟ         ڟ                        0         7  O         V  n         u                        ˨         Ҩ             	           (                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               (!         @!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   h      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-O<-x    O<-J-|    J-7~-\    7~--    --h    -?b-    ?b--	    --h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               @!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        1         2  J         K  c         d  |                                    #          K  T          |                                    +  G          j                         	         
  6          7   c!         d!  "         "  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       :         <  X         Z  v         x                                              
         
  *         ,  H         J  f         h                                                                    8         :  V         X  t         v                                              
           (         *  F         H  d          f   !         #  $         $  &         &  :'         <'  Z(         \(  z)         |)  *         *  +         +  ,         ,  -         -  /         /  :0         <0  Z1         \1  z2         |2  3         3  4         4  5         5  6         6  8         8  :9         <9  Z:         \:  z;         |;  <         <  =         =  >         >  ?         ?  A         A  :B         <B  ZC         \C  zD         |D  E         F  G         G  H         H  J         J  ?K         @K  eL         fL  M         M  N         N  O         O  P         P  #R         $R  ES         GS  hT         jT  U         U  V         V  W         W  X         X  Z         Z  :[         <[  ]\         _\  ]         ]  ^         ^  _         _  `         `  b         b  %c         'c  Fd         Hd  ge         ie  f         f  g         g  h                      ܃              0          X  n                               >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               $!         .!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   C      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-U-     U--    -Y;-x    Y;-I-|    I-A}-\    A}-+-    +--h    -Qb-    Qb-9-	    9--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               '!         6!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         ;  S         T  l         m  	         	  
         
                                                         3          4            O  5          j  P            k                                              9  L           x   !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       5         8  Q         T  m         p                                              	         	             1         4  M         P  i         l                                                                    -         0  I         L  e         h                                                                    )          ,   E!         !  "         "  #         #  %         %  6&         <&  U'         ['  t(         z(  )         )  *         *  +         +  ,         ,  .         .  ./         4/  M0         S0  l1         r1  2         2  3         3  4         4  5         5  7         7  &8         ,8  E9         K9  d:         j:  ;         ;  <         <  =         =  >         >  ?         @  A         $A  =B         CB  \C         C  E         E  0F         6F  OG         UG  nH         tH  I          h  i          j  )k          Sk  il          l  m          m  n         (  A         G  `         f                                 Õ  ܖ                                  9         ?  X         ^  w         }                        ԟ         ڟ                        1         7  P         V  o         u                        ̨         Ҩ             
           )                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               &!         <!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   i      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     --    -O<-x    O<-J-|    J-7~-\    7~-U-    U--h    -Eb-    Eb--	    --h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               <!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        1         2  J         K  c         d  |                                    $          K  U          |                          	          +  H          j                         
           8          9   f!         g!  "         "  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       ;         <  Y         Z  w         x                                              
         
  +         ,  I         J  g         h                                                                    9         :  W         X  u         v                                                         )         *  G         H  e          f   !         #  $         $  &         &  ;'         <'  [(         \(  {)         |)  *         *  +         +  ,         ,  -         -  /         /  ;0         <0  [1         \1  {2         |2  3         3  4         4  5         5  6         6  8         8  ;9         <9  [:         \:  {;         |;  <         <  =         =  >         >  ?         ?  A         A  ;B         <B  [C         \C  {D         |D  E         F  G         G  H         H  J         J  >K         @K  dL         fL  M         M  N         N  O         O  P         P  "R         $R  FS         GS  iT         jT  U         U  V         V  W         W  X         X  Z         Z  ;[         <[  ^\         _\  ]         ]  ^         ^  _         _  `         `  b         b  &c         'c  Gd         Hd  he         ie  f         f  g         g  h          Ԉ              (          P  f                      ̍                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               &!         2!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   D      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     --    -<-    <-J-    J-~-d    ~-A-    A--p    -ib-    ib--
    --
    -簳-t    簳-3׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               9!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             "         #  E         F  h         i                                               "  4          a  s                                              A  J          r  {                       '         (  @          A   Y!         Z!  r"         s"  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                    <  0          Z  N          x  l         %  1&         2&  ['         \'  (         (  )         )  *         *  ,         ,  --         .-  W.         X.  /         /  0         0  1         1  2          3  )4         *4  S5         T5  }6         ~6  7         7  8         8  9         9  %;         &;  O<         P<  y=         z=  >         >  ?         ?  @         @  !B         "B  KC         LC  uD         vD  E         E  F         F  G         G  I         I  FJ         HJ  pK         rK  L         L  M         M  N         N  P         P  BQ         DQ  lR         nR  S         S  T         T  U         U  W         W  >X         @X  hY         jY  Z         Z  [         [  \         \  ^         ^  :_         <_  d`         f`  a         a  b         b  c         c  e         e  6f         8f  `g         bg  h         h  i         i  j         j  l         
l  2m         4m  \n         ^n  o         o  p         p  q         q  s         s  .t         0t  Xu         Zu  v         v  w         w  x         x   z         z  *{         ,{  T|         V|  ~}         }  ~         ~                        &         (  P         R  z         |             Ά         І             "         $  L         N  v         x             ʍ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               1!         J!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    `
         a
             )      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     H
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-:-x    :-I-|    I-|-\    |-7-    7--P    -ub-    ub-c-	    c-g-D
    g-3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               =!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      &         '  ?	         @	  X
         Y
  q         ?  W         X  p         q                                                                      7  1          `  Z                                                 %!          M!  V"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p                        ?         @  _         `                                                        	  
          
  ?         @  _         `                                                                      ?         @  _         `                                                                      ?         @  _         `                          !         !  "         "  #          $  %          %  >&         @&  ^'         `'  ~(         (  )         )  *         *  +         +  ,          -  .          .  >/         @/  ^0         `0  ~1         1  2         2  3         3  4         4  5          6  7          7  >8         @8  ^9         `9  ~:         :  ;         ;  <         <  =         =  >          ?  @          @  >A         G  
I         I  *J         +J  JK         KK  jL         kL  M         M  N         N  O         O  P         P  
R         R  *S         +S  JT         KT  jU         kU  V         V  W         W  X         X  Y         Y  
[         [  *\         +\  J]         K]  j^         k^  _         _  `         `  a         a  b         b  d         d  6e         7e  Xf         Yf  zg         {g  h          tk  l          l  m          Pp  ]q          q  r          r  s                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               9!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 {
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -#-    #-%<-    %<-J-    J-~-d    ~-A-    A--p    -ib-    ib--
    --
    --t    -3׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               =!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             #         $  G         H  k         l                                               '  :          f  y                                              F  P          w                         ,         -  E          F   ^!         _!  w"         x"  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                               	          
           %  0&         2&  Z'         \'  (         (  )         )  *         *  ,         ,  ,-         .-  V.         X.  /         /  0         0  1         1  2          3  (4         *4  R5         T5  |6         ~6  7         7  8         8  9         9  $;         &;  N<         P<  x=         z=  >         >  ?         ?  @         @   B         "B  JC         LC  tD         vD  E         E  F         F  G         G  I         I  GJ         HJ  qK         rK  L         L  M         M  N         N  P         P  CQ         DQ  mR         nR  S         S  T         T  U         U  W         W  ?X         @X  iY         jY  Z         Z  [         [  \         \  ^         ^  ;_         <_  e`         f`  a         a  b         b  c         c  e         e  7f         8f  ag         bg  h         h  i         i  j         j  	l         
l  3m         4m  ]n         ^n  o         o  p         p  q         q  s         s  /t         0t  Yu         Zu  v         v  w         w  x         x  z         z  +{         ,{  U|         V|  }         }  ~         ~                        '         (  Q         R  {         |             φ         І             #         $  M         N  w         x             ˍ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               1!         J!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    `
         a
             *      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     I
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     --    -:-x    :-I-|    I-|-\    |-1-    1--P    -ub-    ub-c-	    c-g-D
    g-3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               :!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      &         '  ?	         @	  X
         Y
  q         ?  W         X  p         q                                                                      7  0          `  Y                                                 &!          M!  W"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p                        >         @  ^         `  ~                                                      	  
          
  >         @  ^         `  ~                                                                    >         @  ^         `  ~                                                                    >         @  ^         `  ~                        !         !  "         "  #          $  %          %  ?&         @&  _'         `'  (         (  )         )  *         *  +         +  ,          -  .          .  ?/         @/  _0         `0  1         1  2         2  3         3  4         4  5          6  7          7  ?8         @8  _9         `9  :         :  ;         ;  <         <  =         =  >          ?  @          @  ?A         G  	I         I  )J         +J  IK         KK  iL         kL  M         M  N         N  O         O  P         P  	R         R  )S         +S  IT         KT  iU         kU  V         V  W         W  X         X  Y         Y  	[         [  )\         +\  I]         K]  i^         k^  _         _  `         `  a         a  b         b  d         d  5e         7e  Wf         Yf  yg         {g  h          s  t          u  %v          Jv  Ww          |w  x          x  y                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               9!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 {
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    <     -,-p     ,--     -#-    #-A5-    A5-C-    C-)w-t    )w-12-     12-9-X    9--    -{b-p	    {b--x
    --
    -s-    s-ճ-\    ճ--<                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ;!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P              	          1  :          b  k                                                          G  H	          p	  q
          
                                              %         &  >           +         ,  K         L  k         l                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        $  %         K  L         L  N         N  EO         FO  mP         nP  Q         Q  R         R  S         S  U         U  5V         6V  ]W         ^W  X         X  Y         Y  Z         Z  [         [  %]         &]  M^         N^  u_         v_  `         `  a         a  b         b  d         d  =e         >e  ef         ff  g         g  h         h  i         i  k         k  -l         .l  Um         Vm  }n         ~n  o         o  p         p  q         q  s         s  Et         Ft  mu         nu  v         v  w         w  x         x  z         z  5{         6{  ]|         ^|  }         }  ~         ~                        %         &  M         N  u         v             ņ         Ɔ                        =         >  e         f                        ݎ         ގ             -         .  T         V  |         ~             ̕         Ε                        D         F  l         n                                                            6  9         ^                        Ԥ         ֤             $         &  L         N  t         v             ī         ƫ                        <         >  d         f                                           b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               <!         L!  1         1  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    `
         a
             )                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    `
         a
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     p
            8     -,-h     ,--     --    -<-    <-K-    K-m~-l    m~--    -G-P    G--    -b-h	    b-!-p
    !-/-
    /-簳-    簳-9׳-L                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               :!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        1         2  J         K  c         d  |                                    #          K  T          |                                    +  B          j                                    M          N   !         !  "         "  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        $  8%         :%  [&         \&  }'         ~'  (         (  )         )  *         *  ,         ,  '-         (-  I.         J.  k/         l/  0         0  1         1  2         2  3         3  5         5  76         86  Y7         Z7  {8         |8  9         9  :         :  ;         ;  =         =  %>         &>  G?         H?  i@         j@  A         A  B         B  C         C  D         D  F         F  5G         6G  VH         XH  xI         zI  J         J  K         K  L         L  "N         $N  UO         WO  P         P  Q         Q  R         R  !T         #T  TU         VU  V         V  W         W  X         X  'Z         (Z  ][         ^[  \         \  ]         ]  ^          _  5`         6`  ka         la  b         b  c         c  e         e  ?f         @f  sg         tg  h         h  i         i  k         k  Cl         Dl  wm         xm  n         n  o         o   q         q  "r         $r  Ds         Fs  ft         ht  u          u  v          v  w          w  x          x  y          z  {         4{  T|         V|  v}         x}  ~         ~             ܀         ހ                        "  B         D  d         f                        ʈ         ̈                        0                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               F!         Z!  1         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    `
         a
             H                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     ?
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    <     -,-p     ,--     -/-    /-;5-    ;5-C-    C-#w-t    #w-12-     12-9-X    9--    -b-p	    b--x
    --
    -s-    s-ճ-\    ճ--<                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ?!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P              
          1  ;          b  l                                                           G  I	          p	  r
          
                                              %         &  >           *         +  I         J  h         i                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        $  %         K  L         L  N         N  DO         FO  lP         nP  Q         Q  R         R  S         S  U         U  4V         6V  \W         ^W  X         X  Y         Y  Z         Z  [         [  $]         &]  L^         N^  t_         v_  `         `  a         a  b         b  d         d  <e         >e  df         ff  g         g  h         h  i         i  k         k  ,l         .l  Tm         Vm  |n         ~n  o         o  p         p  q         q  s         s  Dt         Ft  lu         nu  v         v  w         w  x         x  z         z  4{         6{  \|         ^|  }         }  ~         ~                        $         &  L         N  t         v             Ć         Ɔ                        <         >  d         f                        ܎         ގ             ,         .  U         V  }         ~             ͕         Ε                        E         F  m         n                                              5         6  ]          ^  a                                  ֤  ٥                     &  M         N  u         v             ū         ƫ                        =         >  e         f                                           b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               >!         L!  1         1  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    `
         a
             *                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    `
         a
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     o
            8     -,-h     ,-y-     y--    -<-    <-K-    K-m~-l    m~--    -M-P    M--    -b-h	    b-!-p
    !-/-
    /-簳-    簳-9׳-L                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               8!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        1         2  J         K  c         d  |                                    $          K  U          |                                    +  C          j                                    P          Q   !         !  "         "  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        $  9%         :%  Z&         \&  |'         ~'  (         (  )         )  *         *  ,         ,  &-         (-  H.         J.  j/         l/  0         0  1         1  2         2  3         3  5         5  66         86  X7         Z7  z8         |8  9         9  :         :  ;         ;  =         =  $>         &>  F?         H?  h@         j@  A         A  B         B  C         C  D         D  F         F  4G         6G  WH         XH  yI         zI  J         J  K         K  L         L  #N         $N  VO         WO  P         P  Q         Q  R         R  "T         #T  UU         VU  V         V  W         W  X         X  &Z         (Z  \[         ^[  \         \  ]         ]  ^          _  4`         6`  ja         la  b         b  c         c  
e         e  >f         @f  rg         tg  h         h  i         i  k         k  Bl         Dl  vm         xm  n         n  o         o  q         q  #r         $r  Es         Fs  gt         ht  u         u  v         v  w         w  x         x  z         z  3{          4{  1|          V|  S}          x}  u~          ~                       ހ              !         "  C         D  e         f                        ˈ         ̈                        1                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               D!         Z!  1         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    `
         a
             H                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     @
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -A-    A-7<-x    7<-J-|    J-~-\    ~-4-    4--h    -b-    b-n-	    n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               K!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             %         &  K         L  q         r                                               1  8          p  w                                    (          P  Y                                   6         7  O          P   h!         i!  "         "  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            $         &  J         L  p         r                                            
  .	         0	  T
         V
  z         |                                              8         :  ^         `                                                         B         D  h         j                                               &          (   L!         N!  r"          <  >          (>  ;?          `?  s@          @  A          A  B         K  L         L  M          N  %O         &O  KP         LP  qQ         rQ  R         R  S         S  T         T  	V         
V  /W         0W  UX         VX  {Y         |Y  Z         Z  [         [  \         \  ^         ^  9_         :_  _`         ``  a         a  b         b  c         c  d         d  f         f  Cg         Dg  ih         jh  i         i  j         j  k         k  m         m  'n         (n  Mo           ?         @  e         f                        י         ؙ             #         $  I         J  o         p                                              -         .  S         T  y         z             Ũ         ƨ                        7         8  ]         ^                        ϰ         а                        A         B  g                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               L!         R!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     1
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     -5-    5-M5-    M5-C-    C-5w-l    5w-5-    5--x    -b-	    b--
    --
    -y-|    y-ճ-    ճ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               E!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        ?  S          ~                                    ;  &          V  A          q  \	          	  w
          
                                   0         1  U         V  z         4  L         M  e         f  ~                                                    b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                             <  8          `  \         8"  Y#         [#  |$         ~$  %         %  &         &  '         '  )         )  0*         2*  T+         V+  x,         z,  -         -  .         .  /         /  1         
1  ,2         .2  P3         R3  t4         v4  5         5  6         6  7         7  9         9  (:         *:  L;         N;  p<         r<  =         =  >         >  ?         ?   A         A  $B         &B  HC         JC  lD         nD  E         E  F         F  G         -l  Nm         Pm  qn         sn  o         o  p         p  q         q  s         s  %t         't  Iu         Ku  mv         ov  w         w  x         x  y         y  z         z  !|         #|  E}         G}  i~         k~                        Ձ         ׁ                        A         C  e         g                        щ         Ӊ           ړ                       "  D         F  h         j                        ԛ         ֛                        @         B  d         f                        У         ң                        <         >  `         b                        ̫         Ϋ                        8         :  \         ^                        ȳ         ʳ                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               D!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    `
         a
             .      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    `
         a
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     _
            0     -,-X     ,--     -A-    A-7<-x    7<-J-|    J-~-\    ~-4-    4--h    -b-    b-n-	    n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               K!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             $         %  I         J  n         o                                               ,  4          k  s                                    $          K  U          |                         1         2  J          K   c!         d!  |"         }"  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            %         &  K         L  q         r                                   	         
  /	         0	  U
         V
  {         |                                              9         :  _         `                                                         C         D  i         j                                              '          (   M!         N!  s"          C  D          @D  SE          xE  F          F  G          G  H         K  L         L  M          N  $O         &O  JP         LP  pQ         rQ  R         R  S         S  T         T  V         
V  .W         0W  TX         VX  zY         |Y  Z         Z  [         [  \         \  ^         ^  8_         :_  ^`         ``  a         a  b         b  c         c  d         d  f         f  Bg         Dg  hh         jh  i         i  j         j  k         k   m         m  &n         (n  Lo           >         @  d         f                        ֙         ؙ             "         $  H         J  n         p                                              ,         .  R         T  x         z             Ĩ         ƨ                        6         8  \         ^                        ΰ         а                        @         B  f                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               N!         V!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     0
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     -5-    5-Y5-    Y5-C-    C-Aw-l    Aw-5-    5--x    -b-	    b--
    --
    -y-|    y-ճ-    ճ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               E!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        ?  T          ~                                    ;  %          V  @          q  [	          	  v
          
                                   3         4  Y         Z           >  V         W  o         p                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                                          !  "         8"  Z#         [#  }$         ~$  %         %  &         &  '         '  )         )  1*         2*  U+         V+  y,         z,  -         -  .         .  /         /  	1         
1  -2         .2  Q3         R3  u4         v4  5         5  6         6  7         7  9         9  ):         *:  M;         N;  q<         r<  =         =  >         >  ?         ?  A         A  %B         &B  IC         JC  mD         nD  E         E  F         F  G         -l  Om         Pm  rn         sn  o         o  p         p  q         q  s         s  &t         't  Ju         Ku  nv         ov  w         w  x         x  y         y  z         z  "|         #|  F}         G}  j~         k~                        ց         ׁ                        B         C  f         g                        ҉         Ӊ           ړ             !         "  E         F  i         j                        ՛         ֛                        A         B  e         f                        ѣ         ң                        =         >  a         b                        ͫ         Ϋ                        9         :  ]         ^                        ɳ         ʳ                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               B!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    `
         a
             /      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    `
         a
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     `
            0     -,-X     ,--     -5-    5-#5-x    #5-C-|    C-w-\    w--    --h    -b-    b-!n-	    !n--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               G!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P              	          1  :          b  k                                                          G  C	          p	  l
          
                                              %         &  >           $         %  =         >  V         W  o         p                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            "         $  F         H  j         l                                              	          	  B
         D
  f         h                                                         >         @  b         d                                                         :         <  ^         `                                              !         !  6"          T#  d$          $  %          %  &          &  (          4(  D)         .I  PJ         RJ  tK         vK  L         L  M         M  N         N  P         P  (Q         *Q  LR         NR  pS         rS  T         T  U         U  V         V   X         X  $Y         &Y  HZ         JZ  l[         n[  \         \  ]         ]  ^         ^  _         _  #a         $a  Ib         Jb  oc         pc  d         d  e         e  f         f  h         h  -i         .i  Sj         Tj  yk         m  n         n  p         p  <q         =q  `r         ar  s         s  t         t  u         u  v         v  x         x  8y         9y  \z         ]z  {         {  |         |  }         }  ~         ~             4         5  X         Y  |         }             ą         Ņ                        0         1  T         U  x         y                                                      >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               C!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     D
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-1<-x    1<-J-|    J-~-\    ~--    --h    -b-    b--n-	    -n--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               D!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        1         2  J         K  c         d  |                                    #          K  T          |                                     +  ?          j  ~                                  &          '   K!         L!  p"         q"  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            $         &  J         L  p         r                                            
  .	         0	  T
         V
  z         |                                              8         :  ^         `                                                         B         D  h         j                                               &          (   L!         N!  r"         t"  #         #  $         $  %         %  '         '  1(         2(  W)         X)  }*         ~*  +         +  ,         ,  -         -  /         /  ;0         <0  a1         b1  2          2  3          3  4          5  &6          Q6  i7          7  8         _j  k         k  l         l  m         m  n         n  p         p  Aq         Cq  gr         ir  s         s  t         t  u         u  v         w  %x         'x  Ky         My  qz         sz  {         {  |         |  }         }  	           /         1  U         W  {         }             Ǆ         Ʉ                        9         ;  _         a                        ь         ӌ                      	  .         /  T         U  z         {             ƕ         Ǖ                        8         9  ^         _                        Н         ѝ                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               E!         R!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     O
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-#5-x    #5-C-|    C-w-\    w--    --h    -b-    b-!n-	    !n--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               G!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P              
          1  ;          b  l                                                          G  B	          p	  k
          
                                              %         &  >           $         %  =         >  V         W  o         p                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            #         $  G         H  k         l                                              	          	  C
         D
  g         h                                                         ?         @  c         d                                                         ;         <  _         `                                              !         !  7"          l)  |*          *  +          +  ,          -  $.          L.  \/         .I  QJ         RJ  uK         vK  L         L  M         M  N         N  P         P  )Q         *Q  MR         NR  qS         rS  T         T  U         U  V         V  X         X  %Y         &Y  IZ         JZ  m[         n[  \         \  ]         ]  ^         ^  _         _  "a         $a  Hb         Jb  nc         pc  d         d  e         e  f         f  h         h  ,i         .i  Rj         Tj  xk         m  n         n  p         p  ;q         =q  _r         ar  s         s  t         t  u         u  v         v  x         x  7y         9y  [z         ]z  {         {  |         |  }         }  ~         ~             3         5  W         Y  {         }             Å         Ņ                        /         1  S         U  w         y                                                      >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               C!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     C
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-7<-x    7<-J-|    J-~-\    ~--    --h    -b-    b--n-	    -n--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               F!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        1         2  J         K  c         d  |                                    $          K  U          |                                    +  @          j                                    (          )   N!         O!  t"         u"  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            %         &  K         L  q         r                                   	         
  /	         0	  U
         V
  {         |                                              9         :  _         `                                                         C         D  i         j                                              '          (   M!         N!  s"         t"  #         #  $         $  %         %  
'         '  0(         2(  V)         X)  |*         ~*  +         +  ,         ,  -         -  /         /  :0         <0  `1         b1  2          8  9          :  2;          ];  u<          <  =          =  >         _j  k         k  l         l  m         m  n         n  p         p  Bq         Cq  hr         ir  s         s  t         t  u         u   w         w  &x         'x  Ly         My  rz         sz  {         {  |         |  }         }  
           0         1  V         W  |         }             Ȅ         Ʉ                        :         ;  `         a                        Ҍ         ӌ                      	  -         /  S         U  y         {             ŕ         Ǖ                        7         9  ]         _                        ϝ         ѝ                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               E!         R!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     N
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-_;-x    _;-I-|    I-G}-\    G}-+-    +--h    -{b-    {b-x-	    x--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               E!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         ;  S         T  l         m  	         	  
         
           S  }         ~                                   )          *  !          E  <          `  W          {  r                                    
          /  I           n   !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                      "  B         D  d         f                                              	         	  0
         2
  R         T  t         v                                                          @         B  b         d                                                         .         0  P         R  r         t                                     !         "  1#         3#  S$         sG  H         H  I         I  J         J  K         K  M         M  =N         ?N  _O         aO  P         P  Q         Q  R         R  S         S  	U         U  +V         -V  MW         OW  oX         qX  Y         Y  Z         Z  [         [  \         \  ^         ^  ;_         =_  ]`         _`  a         a  b         b  c         c  d         d  f         	f  )g         +g  Kh         Mh  mi         oi  j         j  k         k  l          9y  8z          ]z  \{          {  |          |  }          }  ~         $  D         F  f         h                        ̕         Ε                        2         4  T         V  v         x                        ܞ         ޞ                        "  B         D  d         f                        ʦ         ̦                        0         2  R         T  t                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               >!         J!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 w
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-G5-x    G5-C-|    C-/w-\    /w-C-    C--h    -b-    b-'n-	    'n--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               D!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        ?  S          ~                                    ;  &          V  A          q  \	          	  w
          
                                   0         1  U         V  z         4  L         M  e         f  ~                                                    b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            &         (  N         P  v         x                                              >	         @	  f
         h
                                              .         0  V         X  ~                                                        F         H  n         p                                              6         8  ^          `   !         !  "         "  $         $  :%         <%  b&         d&  '         '  (         (  )         )  +         +  *,         ,,  R-         T-  z.         |.  /         /  0         0  1         1  3         3  B4         D4  j5         l5  6         6  7         7  8         8  
:         :  2;         4;  Z<         \<  =         =  >         >  ?         ?  @         @  "B         $B  JC         LC  rD         tD  E         E  F          T  U          U  V          
W  X          FX  WY          Y  Z         Bm  hn         jn  o         o  p         p  q         q  s         
s  0t         2t  Xu         Zu  v         v  w         w  x         x  y         y   {         "{  H|         J|  p}         r}  ~         ~                                   8         :  `         b                        ؇         ڇ              (         *  P         R  x         z             Ȏ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               D!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     O
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-k;-x    k;-J-|    J-S}-\    S}-+-    +--h    -b-    b-x-	    x--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               A!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         ;  S         T  l         m  	         	  
         
           X                                              3          4  *          O  E          j  `            {                                              9  T           x   !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            !         "  C         D  e         f                                              	         	  1
         2
  S         T  u         v                                                          A         B  c         d                                                         /         0  Q         R  s         t                                     !         "  2#         3#  T$         sG  H         H  I         I  J         J  K         K  M         M  >N         ?N  `O         aO  P         P  Q         Q  R         R  S         S  
U         U  ,V         -V  NW         OW  pX         qX  Y         Y  Z         Z  [         [  \         \  ^         ^  <_         =_  ^`         _`  a         a  b         b  c         c  d         d  f         	f  *g         +g  Lh         Mh  ni         oi  j         j  k         k  l          ~                        5  4          Y  X          }  |         $  E         F  g         h                        ͕         Ε                        3         4  U         V  w         x                        ݞ         ޞ              !         "  C         D  e         f                        ˦         ̦                        1         2  S         T  u                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               @!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 w
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-S5-x    S5-C-|    C-;w-\    ;w-C-    C--h    -b-    b-'n-	    'n--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               F!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        ?  T          ~                                    ;  %          V  @          q  [	          	  v
          
                                   3         4  Y         Z           >  V         W  o         p                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            '         (  O         P  w         x                                              ?	         @	  g
         h
                                              /         0  W         X                                                          G         H  o         p                                              7         8  _          `   !         !  "         "  $         $  ;%         <%  c&         d&  '         '  (         (  )         )  +         +  +,         ,,  S-         T-  {.         |.  /         /  0         0  1         1  3         3  C4         D4  k5         l5  6         6  7         7  8         8  :         :  3;         4;  [<         \<  =         =  >         >  ?         ?  @         @  #B         $B  KC         LC  sD         tD  E         E  F          Z  [          [  ]          6]  G^          r^  _          _  `         Bm  in         jn  o         o  p         p  q         q  	s         
s  1t         2t  Yu         Zu  v         v  w         w  x         x  y         y  !{         "{  I|         J|  q}         r}  ~         ~                                   9         :  a         b                        ه         ڇ             )         *  Q         R  y         z             Ɏ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               B!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     N
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-5-x    5-C-|    C-w-\    w--    --h    -cb-    cb-x-	    x--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               E!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P              	          1  :          b  k                                                          G  C	          p	  l
          
                                              %         &  >           $         %  =         >  V         W  o         p                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         X  H          v  f                                             "  $         $  *%         ,%  J&         L&  j'         l'  (         (  )         )  *         *  +         +  
-         -  *.         ,.  J/         L/  j0         l0  1         1  2         2  3         3  4         4  
6         6  *7         ,7  J8         L8  j9         l9  :         :  ;         ;  <         <  =         =  
?         ?  *@         ,@  JA         LA  jB         lB  C         C  D         D  E         AF  _G         aG  H         H  I         I  J         J  K         K  L         M  N         !N  ?O         AO  _P         aP  Q         Q  R         R  S         S  T         T  U         V  W         !W  ?X         AX  _Y         aY  Z         Z  [         [  \         \  ]         ]  ^         _  `         !`  ?a         Aa  _b         ab  c         c  d         d  e         e  f         f  g         h  i         i  k         k  1l         2l  Qm         Rm  qn         rn  o         o  p         p  q         q  r         r  t         t  1u         2u  Qv         Rv  qw         rw  x         x  y         y  z         z  {         {  }         }  1~         2~  Q         R  q         r                        у         ҃                        1         2  Q         R  q                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               5!         <!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 w
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-:-x    :-I-|    I-|-\    |-i-    i--h    -Qb-    Qb-x-	    x--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               F!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      &         '  ?	         @	  X
         Y
  q         ?  W         X  p         q                                                                      7  9          `  b                                                 %!          M!  V"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            )         *  S         T  }         ~                                   %         &  O	         P	  y
         z
                                   !         "  K         L  u         v                                              G         H  q         r                                              C         D  m         n               !         !  "         $  <%         >%  f&         h&  '         '  (         (  )         )  +         +  8,         :,  b-         d-  .         .  /         /  0         0  
2         2  43         63  ^4         `4  5         5  6         6  7         7  9         9  0:         2:  Z;         \;  <         <  =         =  >         >  @         @  ,A         .A  VB         XB  C         C  D         D  E         E  F          G  (H         :H  bI         dI  J         J  K         K  L         L  
N         N  4O         6O  ^P         `P  Q         Q  R         R  S         S  U         U  0V         2V  ZW         \W  X         X  Y         Y  Z         Z  \         \  ,]         .]  V^         X^  _         _  `         `  a         a  b          c  (d         *d  Re         Te  |f         ~f  g         g  h          2~  *          R  J          r  j                                 ʎ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               '!         :!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-5-x    5-C-|    C-w-\    w--    --h    -ib-    ib-x-	    x--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               A!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P              
          1  ;          b  l                                                          G  B	          p	  k
          
                                              %         &  >           $         %  =         >  V         W  o         p                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                 *            H  8           f   V!         "  
$         $  +%         ,%  K&         L&  k'         l'  (         (  )         )  *         *  +         +  -         -  +.         ,.  K/         L/  k0         l0  1         1  2         2  3         3  4         4  6         6  +7         ,7  K8         L8  k9         l9  :         :  ;         ;  <         <  =         =  ?         ?  +@         ,@  KA         LA  kB         lB  C         C  D         D  E         AF  `G         aG  H         H  I         I  J         J  K         K   M         M   N         !N  @O         AO  `P         aP  Q         Q  R         R  S         S  T         T   V         V   W         !W  @X         AX  `Y         aY  Z         Z  [         [  \         \  ]         ]   _         _   `         !`  @a         Aa  `b         ab  c         c  d         d  e         e  f         f   h         h  i         i  k         k  0l         2l  Pm         Rm  pn         rn  o         o  p         p  q         q  r         r  t         t  0u         2u  Pv         Rv  pw         rw  x         x  y         y  z         z  {         {  }         }  0~         2~  P         R  p         r                        Ѓ         ҃                        0         2  P         R  p                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               7!         @!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 w
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-:-x    :-I-|    I-|-\    |-i-    i--h    -Wb-    Wb-x-	    x--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               D!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      &         '  ?	         @	  X
         Y
  q         ?  W         X  p         q                                                                      7  8          `  a                                                 &!          M!  W"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            (         *  R         T  |         ~                                   $         &  N	         P	  x
         z
                                             "  J         L  t         v                                              F         H  p         r                                              B         D  l         n               !         !  "         $  =%         >%  g&         h&  '         '  (         (  )         )  +         +  9,         :,  c-         d-  .         .  /         /  0         0  2         2  53         63  _4         `4  5         5  6         6  7         7  9         9  1:         2:  [;         \;  <         <  =         =  >         >  @         @  -A         .A  WB         XB  C         C  D         D  E         E  F          G  )H         :H  cI         dI  J         J  K         K  L         L  N         N  5O         6O  _P         `P  Q         Q  R         R  S         S  U         U  1V         2V  [W         \W  X         X  Y         Y  Z         Z  \         \  -]         .]  W^         X^  _         _  `         `  a         a  b          c  )d         *d  Se         Te  }f         ~f  g         g  h          ҃  ʄ                        
          2  *          R  J         ʎ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               +!         >!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-O-     O--    -Y;-x    Y;-I-|    I-A}-\    A}--    --h    -ib-    ib-E-	    E--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               &!         0!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         ;  S         T  l         m  	         	  
         
           S  }         ~                                   )          *  !          E  <          `  W          {  r                                    
          /  I           n   !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       :         <  X         Z  v         x                                              
         
  *         ,  H         J  f         h                                                                    8         :  V         X  t         v                                              
           (         *  F         H  d          f   !         p"  #         #  $         $  %         %  &         &  (         (  ")         $)  @*         B*  ^+         `+  |,         ~,  -         -  .         .  /         /  0         0  2         2  03         23  N4         P4  l5         n5  6         6  7         7  8         8  9         9  ;         ;   <         "<  >=         @=  \>         ^>  z?         |?  @         @  A         A  B         B  C         C  E         E  .F         0F  LG         :H  VI         XI  tJ         vJ  K         K  L         L  M         M  N         N  
P         P  (Q         *Q  FR         HR  dS         fS  T         T  U         U  V         V  W         W  X         X  Z         Z  6[         8[  T\         V\  r]         t]  ^         ^  _         _  `         `  a         a  c         
c  &d         (d  De         Fe  bf          Bm  \n          ~n  o          o  p          p  r          2r  Ls                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               2!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   p      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-y-     y--    -%<-x    %<-J-|    J-~-\    ~--    --h    -9b-    9b--	    --h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               8!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        1         2  J         K  c         d  |                                    #          K  T          |                                     +  ?          j  ~                                  &          '   K!         L!  p"         q"  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       5         :  R         W  o         t                                               
         
           "  :         ?  W         \  t         y                                                       
  "         '  ?         D  \         a  y         ~                                              
           '         ,  D          I   a!         d#  ~$         $  %         %  &         &  '         '  )         )  '*         ,*  K+         P+  o,         t,  -         -  .         .  /         /  0         1  !2         &2  C3         H3  e4         j4  5         5  6         6  7         7  8         8  :         :  1;         6;  S<         X<  s=         x=  >         >  ?         ?  @         @  A         A  C         C  3D         8D  SE         XE  sF         xF  G          I  J          J  L          5L  PM          xM  N          N  O         m  n         n  9o         ;o  Up         Wp  qq         sq  r         r  s         s  t         t  u         u  v         v  x         x  5y         7y  Qz         Sz  m{         o{  |         |  }         }  ~         ~                                   1         3  M         O  i         k                                   ى         ۉ                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               !         ,!  1         1  z=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   C      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-I-     I--    -_;-x    _;-I-|    I-G}-\    G}--    --h    -]b-    ]b-E-	    E--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               #!         .!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         ;  S         T  l         m  	         	  
         
           X                                              3          4  *          O  E          j  `            {                                              9  T           x   !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       ;         <  Y         Z  w         x                                              
         
  +         ,  I         J  g         h                                                                    9         :  W         X  u         v                                                         )         *  G         H  e          f   !         p"  #         #  $         $  %         %  &         &  (         (  #)         $)  A*         B*  _+         `+  },         ~,  -         -  .         .  /         /  0         0  2         2  13         23  O4         P4  m5         n5  6         6  7         7  8         8  9         9  ;         ;  !<         "<  ?=         @=  ]>         ^>  {?         |?  @         @  A         A  B         B  C         C  E         E  /F         0F  MG         :H  WI         XI  uJ         vJ  K         K  L         L  M         M  N         N  P         P  )Q         *Q  GR         HR  eS         fS  T         T  U         U  V         V  W         W  X         X  Z         Z  7[         8[  U\         V\  s]         t]  ^         ^  _         _  `         `  a         a  	c         
c  'd         (d  Ee         Fe  cf          ns  t          t  u          u   w          "w  <x          ^x  xy                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               .!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   q      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-s-     s--    -+<-x    +<-J-|    J-~-\    ~--    --h    -?b-    ?b--	    --h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               6!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        1         2  J         K  c         d  |                                    $          K  U          |                                    +  @          j                                    (          )   N!         O!  t"         u"  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       6         :  S         W  p         t                                              
         
           "  ;         ?  X         \  u         y                                                       
  #         '  @         D  ]         a  z         ~                                                         (         ,  E          I   b!         d#  $         $  %         %  &         &  '         '  )         )  (*         ,*  L+         P+  p,         t,  -         -  .         .  /         /   1         1  "2         &2  D3         H3  f4         j4  5         5  6         6  7         7  8         8  :         :  2;         6;  T<         X<  t=         x=  >         >  ?         ?  @         @  A         A  C         C  4D         8D  TE         XE  tF         xF  G          O  Q          AQ  \R          R  S          S  T          
U  %V         m  n         n  :o         ;o  Vp         Wp  rq         sq  r         r  s         s  t         t  u         u  v         v  x         x  6y         7y  Rz         Sz  n{         o{  |         |  }         }  ~         ~                                   2         3  N         O  j         k                                   ډ         ۉ                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               "!         0!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   D      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-=<-x    =<-J-|    J-%~-\    %~--    --h    -ob-    ob-?-	    ?--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               E!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             *         +  U         V                                                %          J  d                                              8  A          i  r                                 7  O         P  h          i   !         !  "         "  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       8         :  U         W  r         t                                              
         
            "  =         ?  Z         \  w         y                                                       
  %         '  B         D  _         a  |         ~                                                         *         ,  G          I   d!         d#  $         $  %         %  &         &  '         '  )         )  **         ,*  N+         P+  r,         t,  -         -  .         .  /         /  1         1  $2         &2  F3         H3  h4         j4  5         5  6         6  7         7  8         8  :         :  4;         6;  V<         X<  v=         x=  >         >  ?         ?  @         @  A         A  C         C  6D         8D  VE         XE  vF         xF  G          gU  qV          V  W          W  X          X  Y          Z  ![         m  n         n  :o         ;o  Vp         Wp  rq         sq  r         r  s         s  t         t  u         u  v         v  x         x  6y         7y  Rz         Sz  n{         o{  |         |  }         }  ~         ~                                   2         3  N         O  j         k                                   ډ         ۉ                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               7!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   p      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-g-     g--    -:-x    :-{I-|    {I-|-\    |-5-    5--P    -]b-    ]b-c-	    c-g-D
    g--׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               /!         @!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      &         '  ?	         @	  X
         Y
  q         ?  W         X  p         q                                                                      7  9          `  b                                                 %!          M!  V"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p                     "  =         D  _         f                                              		         	  +
         2
  M         T  o         v                                                          ;         B  ]         d                                                         )         0  K         R  m         t                                     !         !  #         #  2$         7$  Q%         V%  p&         u&  '         '  (         (  )         )  +         +  6,         =,  ^-         e-  .         .  /         /  0         0  1         2  (3         /3  R4         Y4  |5         5  6         6  7         7  8         9  $:         +:  N;         U;  x<         <  =         =  >         >  ?         @  &A         -A  RB         YB  ~C         C  D         D  E         E  G         	G  .H         5H  OI          W  X          X  Y           Z  [          @[  [\          \  ]          n  o         o  9p         >p  Xq         ]q  wr         |r  s         s  t         t  u         u  v         v  x         x  1y         6y  Pz         Uz  o{         t{  |         |  }         }  ~         ~             
           )         .  H         M  g         l                        ć                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               1!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 {
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-C<-x    C<-J-|    J-+~-\    +~--    --h    -ob-    ob-?-	    ?--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               G!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             +         ,  W         X                                                +          O  j                                              =  G          n  x                                 <  T         U  m          n   !         !  "         "  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       9         :  V         W  s         t                                              
         
  !         "  >         ?  [         \  x         y                                              	         
  &         '  C         D  `         a  }         ~                                                         +         ,  H          I   e!         d#  ~$         $  %         %  &         &  '         '  )         )  +*         ,*  O+         P+  s,         t,  -         -  .         .  /         /  1         1  %2         &2  G3         H3  i4         j4  5         5  6         6  7         7  8         8  :         :  5;         6;  W<         X<  w=         x=  >         >  ?         ?  @         @  A         A  C         C  7D         8D  WE         XE  wF         xF  G          C[  M\          o\  y]          ]  ^          ^  _          _  `         m  n         n  9o         ;o  Up         Wp  qq         sq  r         r  s         s  t         t  u         u  v         v  x         x  5y         7y  Qz         Sz  m{         o{  |         |  }         }  ~         ~                                   1         3  M         O  i         k                                   ى         ۉ                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               7!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   q      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-a-     a--    -:-x    :-{I-|    {I-|-\    |-5-    5--P    -]b-    ]b-c-	    c-g-D
    g--׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               .!         >!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      &         '  ?	         @	  X
         Y
  q         ?  W         X  p         q                                                                      7  8          `  a                                                 &!          M!  W"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p                     "  <         D  ^         f                                              	         	  *
         2
  L         T  n         v                                                          :         B  \         d  ~                                                       (         0  J         R  l         t                                     !         !  #         #  3$         7$  R%         V%  q&         u&  '         '  (         (  )         )  +         +  5,         =,  ]-         e-  .         .  /         /  0         0  1         2  '3         /3  Q4         Y4  {5         5  6         6  7         7  8         9  #:         +:  M;         U;  w<         <  =         =  >         >  ?         @  %A         -A  QB         YB  }C         C  D         D  E         E  G         	G  -H         5H  PI          ]  ^           _  `          @`  [a          a  b          b  c          n  o         o  :p         >p  Yq         ]q  xr         |r  s         s  t         t  u         u  v         v  x         x  2y         6y  Qz         Uz  p{         t{  |         |  }         }  ~         ~                        *         .  I         M  h         l                        Ň                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               1!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 {
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -#-    #-M;-    M;-I-    I-5}-d    5}-5-    5--0    --    -b-H	    b-c-P
    c--
    -'׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               B!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         ;  S         T  l         m  	         	  
         
           S  }         ~                                   )          *  !          E  <          `  W          {  r                                    
          /  I           n   !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q   w  operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       \                      "  B         D  d         f                                              	         	  0
         2
  R         T  t         v                                                          @         B  b         d                                                         .         0  P         R  r         t                                     !         !  #         #  5$         7$  T%         V%  s&         u&  '         '  (         (  )         )  +         +  ;,         =,  c-         e-  .         .  /         /  0         0  2         2  -3         /3  W4         Y4  5         5  6         6  7         7  8         9  ):         +:  S;         U;  }<         <  =         =  >         >  ?         @  +A         -A  WB         YB  C         C  D         D  E         E  G         	G  3H         5H  RI          n  o         o  <p         >p  [q         ]q  zr         |r  s         s  t         t  u         u  v         v  x         x  4y         6y  Sz         Uz  r{         t{  |         |  }          }  ~          ~                          	          .  (         M  j         l                        Ǉ                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ^!         l!  2         "2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,-[-     [--    -<-    <-J-    J-~-d    ~-	-    	--p    -b-    b--
    --
    -հ-t    հ-'׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               .!         >!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        1         2  J         K  c         d  |                                    #          K  T          |                                     +  ?          j  ~                                  &          '   K!         L!  p"         q"  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                    +          P  [                     #  $         $  %         %  &         '  #(         *(  G)         N)  k*         r*  +         +  ,         ,  -         -  .         /  0         &0  C1         J1  g2         n2  3         3  4         4  5         5  6         6  8         "8  ?9         F9  c:         j:  ;         ;  <         <  =         =  >         >  @         @  ;A         BA  _B         fB  C         C  D         D  E         E  F         F  H         H  6I         ;I  WJ         \J  xK         }K  L         L  M         M  N         N  P         
P  ,Q         .Q  PR         RR  tS         vS  T         T  U         U  V         V  X         X  &Y         (Y  HZ         JZ  j[         l[  \         \  ]         ]  ^         ^  _         _  a         a  6b         8b  Xc         Zc  xd         zd  e         e  f         f  g         g  h         h  j         j  8k         :k  Xl         Zl  xm         zm  n         n  o         o  p         p  q         q  s         s  :t         ?t  [u         `u  |v         v  w         w  x         x  y         y   {         {  !|         &|  B}         G}  c~         h~                        Ɓ         ˁ                        )         .  J         O  k         p                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               H!         R!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    `
         a
             .      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     ?
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -] -`     ] -հ-     հ-%ܰ-    %ܰ-]4-x    ]4-@-\    @-h-<    h--    -;-    ;--    -b- 	    b-c-(
    c--
    -'׳-
                   =         b      J      7   U   7   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               E!         R!  R1                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 64 })  outputElements  5184    input0Elements  31104   input1Elements  24576   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     4                           #         b      J      7   M     operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       @                      -         .  F         G  _	                      7         8  c         d                                                                        ;  V          z                                            b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     4                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  1728    input0Elements  1728    invocations     1                           3         b      J      7   Q   w  operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       \            !         "  C         D  e         f                                              	         	  1
         2
  S         T  u         v                                                          A         B  c         d                                                         /         0  Q         R  s         t                                     !         !  #         #  6$         7$  U%         V%  t&         u&  '         '  (         (  )         )  +         +  <,         =,  d-         e-  .         .  /         /  0         0  2         2  .3         /3  X4         Y4  5         5  6         6  7         7   9         9  *:         +:  T;         U;  ~<         <  =         =  >         >   @         @  ,A         -A  XB         YB  C         C  D         D  E         E  G         	G  4H         5H  SI          n  o         o  =p         >p  \q         ]q  {r         |r  s         s  t         t  u         u  v         v  x         x  5y         6y  Tz         Uz  s{         t{  |         |  }         }  ~         ~                        -         .  L          M  G          l  f                                                      b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     4                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               \!         h!  2         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,-g-     g--    -<-    <-J-    J-~-d    ~-	-    	--p    -{b-    {b--
    --
    -հ-t    հ-'׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               2!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        1         2  J         K  c         d  |                                    $          K  U          |                                    +  @          j                                    (          )   N!         O!  t"         u"  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                    !          @!  K"          p"  {#         #  $         $  %         %  &         '  "(         *(  F)         N)  j*         r*  +         +  ,         ,  -         -  .         /  0         &0  B1         J1  f2         n2  3         3  4         4  5         5  6         6  8         "8  >9         F9  b:         j:  ;         ;  <         <  =         =  >         >  @         @  :A         BA  ^B         fB  C         C  D         D  E         E  F         F  H         H  7I         ;I  XJ         \J  yK         }K  L         L  M         M  N         N  	P         
P  -Q         .Q  QR         RR  uS         vS  T         T  U         U  V         V  X         X  'Y         (Y  IZ         JZ  k[         l[  \         \  ]         ]  ^         ^  _         _  a         a  7b         8b  Yc         Zc  yd         zd  e         e  f         f  g         g  h         h  j         j  9k         :k  Yl         Zl  ym         zm  n         n  o         o  p         p  q         q  s         s  ;t         ?t  \u         `u  }v         v  w         w  x         x  y         y  {         {  "|         &|  C}         G}  d~         h~                        ǁ         ˁ             	           *         .  K         O  l         p                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               F!         R!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    `
         a
             /      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     @
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-7<-x    7<-J-|    J-~-\    ~-36-    36--h    -b-    b-	n-	    	n--h
    -!׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               D!         X!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             *         +  U         V                                                %          J  d                                              8  A          i  r                                 7  O         P  h          i   !         !  "         "  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            (         *  R         T  |         ~                                   $         &  N	         P	  x
         z
                                             "  J         L  t         v                                              F         H  p         r                                              B         D  l         n               !         !  "          %  &          D&  ['          '  (          (  )          )  +         K  L         L  N         N  1O         2O  [P         \P  Q         Q  R         R  S         S  U         U  -V         .V  WW         XW  X         X  Y         Y  Z         Z  [          \  )]         *]  S^         T^  }_         ~_  `         `  a         a  b         b  %d         &d  Oe         Pe  yf         zf  g         g  h         h  i         i  !k         "k  Kl         Ll  um         vm  n         n  o           6         8  `         b                        ޚ                    
  2         4  \         ^                        ڡ         ܡ             .         0  X         Z                        ֨         ب              *         ,  T         V  ~                      ү         ԯ             &         (  P         R  z         |             ζ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ]!         b!  2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     D
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                 w        Sܯ--0     --     -c-     c-B-P    B-c-    c--    -!׳-p                             b      *      7   I      operation.0     sync             :            2            b         reason  ExchangeBarrier value   1       invocations     1                                     b      *      7   I      operation.0     sync             :            2            b         reason  ExchangeBarrier value   3       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   5       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   2       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-=<-x    =<-J-|    J-%~-\    %~-96-    96--h    -b-    b-	n-	    	n--h
    -!׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               F!         X!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             +         ,  W         X                                                +          O  j                                              =  G          n  x                                 <  T         U  m          n   !         !  "         "  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            )         *  S         T  }         ~                                   %         &  O	         P	  y
         z
                                   !         "  K         L  u         v                                              G         H  q         r                                              C         D  m         n               !         !  "          4+  K,          p,  -          -  .          .  /          $0  ;1         K  L         L  N         N  0O         2O  ZP         \P  Q         Q  R         R  S         S  U         U  ,V         .V  VW         XW  X         X  Y         Y  Z         Z  [          \  (]         *]  R^         T^  |_         ~_  `         `  a         a  b         b  $d         &d  Ne         Pe  xf         zf  g         g  h         h  i         i   k         "k  Jl         Ll  tm         vm  n         n  o           7         8  a         b                        ߚ           	         
  3         4  ]         ^                        ۡ         ܡ             /         0  Y         Z                        ר         ب             +         ,  U         V                        ӯ         ԯ             '         (  Q         R  {         |             ϶                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               a!         f!  2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     C
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                 w        Sܯ--0     --     -c-     c-B-P    B-c-    c--    -!׳-p                             b      *      7   I      operation.0     sync             :            2            b         reason  ExchangeBarrier value   1       invocations     1                                     b      *      7   I      operation.0     sync             :            2            b         reason  ExchangeBarrier value   3       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   5       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   2       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -{,-h     {,-1-     1--    -#5-    #5-C-    C-w-l    w-30-    30--x    -cb-	    cb--
    --
    -U-|    U-ճ-    ճ-׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     f            !          B!  1          1  y=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        /  4          ^  c                                                          A  E	          l	  p
          
                                              -         .  H            :         ;  U         V  p         q                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |           +          R  a                                             T$  v%         x%  &         vL  M         M  N         N  O         O  Q         Q  )R         *R  MS         NS  qT         rT  U         U  V         V  W         W  Y         Y  %Z         &Z  I[         J[  m\         n\  ]         ]  ^         ^  _         _  `         `  !b         "b  Ec         Fc  id         jd  e         e  f         f  g         g  h         h  j         j  Ak         Bk  el         fl  m         m  n         n  o         o  p         p  r         r  =s         >s  at         bt  u         u  v         v  w         w  x         x  z         z  9{         :{  ]|         ^|  }         }  ~         ~                                   5         6  Y         Z  }         ~             Ň         Ƈ                        1         2  U         V  y         z                                            
  ,         .  P         R  t         v                                              (         *  L         N  p         r                        ܠ         ޠ              $         &  H         J  l         n                        ب         ڨ                       "  D         F  h         j                        ԰                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               C!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    b
         c
             /      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    b
         c
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,-C-     C--    -<-    <-J-    J-}-l    }-a-    a--P    --    -b-h	    b--p
    --
    -ð-    ð--L                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     q            !          F!  1          1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        5         6  P         Q  k         l                                       %          O  T          ~                          
          +  I          j                                               !   G!         H!  n"         o"  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        $  %         %  &         &  (         (  ')         ()  M*         N*  s+         t+  ,         ,  -         -  .         .  0         0  11         21  W2         X2  }3         ~3  4         4  5         5  6         6  8         8  ;9         <9  a:         b:  ;         ;  <         <  =         =  >         >  @          @  EA         FA  kB         lB  C         C  D         D  E         E  G         G  (H         *H  NI         PI  tJ         vJ  K         K  L         L  M         M  O         O  2P         4P  XQ         ZQ  ~R         R  S         S  T         T  U         U  W         W  <X         >X  bY         dY  Z         Z  [         [  \         \  ]         ]   _         "_  F`         H`  la         na  b         b  c         c  d         d  f         
f  0g         2g  Xh         Zh  i         i  j         j  k         k  l         l   n         "n  Ho         Jo  np         pp  q         q  r         r  s         s  u         u  ,v         .v  Rw         Tw  xx         zx  y         y  z         z  {         {  }         }  6~         8~  \         ^                                    Ђ  σ                                 B  f         h                        ؊         ڊ                                b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ]!         ^!  2         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    b
         c
             4                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     ;
                   
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            8     -{,-h     {,-1-     1--    -#5-    #5-C-    C-w-l    w-30-    30--x    -Wb-	    Wb--
    --
    -U-|    U-ճ-    ճ-׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     e            !          B!  1          1  x=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        /  5          ^  d                                                          A  D	          l	  o
          
                                              -         .  H            :         ;  U         V  p         q                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         *  9          `  o                                             T$  w%         x%  &         vL  M         M  N         N  O         O  Q         Q  (R         *R  LS         NS  pT         rT  U         U  V         V  W         W   Y         Y  $Z         &Z  H[         J[  l\         n\  ]         ]  ^         ^  _         _  `         `   b         "b  Dc         Fc  hd         jd  e         e  f         f  g         g  h         h  j         j  @k         Bk  dl         fl  m         m  n         n  o         o  p         p  r         r  <s         >s  `t         bt  u         u  v         v  w         w  x         x  z         z  8{         :{  \|         ^|  }         }  ~         ~                                   4         6  X         Z  |         ~             ć         Ƈ                        0         2  T         V  x         z                                   	         
  -         .  Q         R  u         v                                              )         *  M         N  q         r                        ݠ         ޠ             %         &  I         J  m         n                        ٨         ڨ             !         "  E         F  i         j                        հ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ?!         @!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    b
         c
             0      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    b
         c
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,-=-     =--    -<-    <-J-    J-~-l    ~-7-    7--P    --    -b-h	    b--p
    --
    -ɰ-    ɰ--L                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     p            !          F!  1          1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        5         6  P         Q  k         l                                       &          O  U          ~                          	          +  H          j                                    "          #   J!         K!  r"         s"  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        $  %         %  &         &   (         (  &)         ()  L*         N*  r+         t+  ,         ,  -         -  .         .  
0         0  01         21  V2         X2  |3         ~3  4         4  5         5  6         6  8         8  :9         <9  `:         b:  ;         ;  <         <  =         =  >         >  @          @  DA         FA  jB         lB  C         C  D         D  E         E  G         G  )H         *H  OI         PI  uJ         vJ  K         K  L         L  M         M  O         O  3P         4P  YQ         ZQ  R         R  S         S  T         T  U         U  W         W  =X         >X  cY         dY  Z         Z  [         [  \         \  ]         ]  !_         "_  G`         H`  ma         na  b         b  c         c  d         d  	f         
f  1g         2g  Yh         Zh  i         i  j         j  k         k  l         l  !n         "n  Io         Jo  op         pp  q         q  r         r  s         s  u         u  -v         .v  Sw         Tw  yx         zx  y         y  z         z  {         {  }         }  7~         8~  ]         ^                        ς         Ђ                        A          B  A          h  g                                  ڊ  ً                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ]!         ^!  2         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    b
         c
             5                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     <
                   
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,-=-     =--    -;-x    ;-}J-|    }J-}-\    }-c3-    c3--h    -ub-    ub-m-	    m--h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     h            !          H!  1          1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        9         :  V         W  s         t            |                                    9  M          x                                                D  I          s  x                                   -!         .!  H"         I"  c#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                      '  G         N  n         u                                   
           1	         8	  X
         _
                                                       "  B         I  i         p                                              ,         3  S         Z  z                                                       =          D   d!         k!  "          ?  /@          Y@  rA          A  B          B  C          "D  ;E         K  L         L  N         N  5O         >O  cP         lP  Q         Q  R         R  S         S  U         $U  IV         RV  wW         W  X         X  Y         Y  Z         [  '\         0\  S]         \]  ^         ^  _         _  `         `  b         b  /c         8c  Yd         bd  e         e  f         f  g         g  i         
i  +j         4j  Uk         ^k  l         l  m         m  n         ȑ                        0         4  T         X  x         |                      ę                        ,         0  P         T  t         x                                              (         ,  L         P  p         t                        ܪ                       $         (  H         L  l         p                        ز         ܲ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               J!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     8
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            8     -,-h     ,-m-     m--    -/5-    /5-C-    C-w-l    w---    ---x    -?b-	    ?b--
    --
    -m-|    m-ճ-    ճ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               9!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        ?  ]          ~                                    ;  2          T  K          m  d	          	  }
          
                                   ,         -  S         T  z         *  D         E  _         `  z         {                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |        V"  {#         }#  $          $  %          %  &          '  (          :(  J)          l)  |*         H  I         I  J         J  K         K  M         M  EN         GN  mO         oO  P         P  Q         Q  R         R  T         T  5U         7U  ]V         _V  W         W  X         X  Y         Y  Z         Z  %\         '\  M]         O]  u^         w^  _         _  `         `  a         a  c         c  =d         ?d  ee         ge  f         f  g         g  h         h  j         j  -k         /k  Ul         Wl  }m         m  n         n  o         o  p         p  r         r  Es         Gs  mt         ot  u         u  v         v  w         w  y         y  5z         7z  ]{         _{  |         |  }         }  ~         ~             %         '  M         O  u         w             Ņ         ǅ                        =         ?  e         g                        ݍ         ߍ             ,         .  S         U  {         }             ˔         ͔                        C         E  k         m                                              3         5  [         ]                        ӣ         գ             #         %  K         M  s         u             ê         Ū                        ;                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     q            !          L!  1          1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    b
         c
             4      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    d
         e
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            0     -,-X     ,-=-     =--    -;-x    ;-J-|    J-}-\    }-c3-    c3--h    -ub-    ub-m-	    m--h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     i            !          H!  1          1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        ;         <  Y         Z  w         x                                                >  Q          }                                                 I  O          x  ~                                   2!         3!  M"         N"  h#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            !         '  H         N  o         u                                              2	         8	  Y
         _
                                                       "  C         I  j         p                                              -         3  T         Z  {                                                       >          D   e!         k!  "          eE  ~F          F  G          G  I          .I  GJ          qJ  K         K  L         L  N         N  4O         >O  bP         lP  Q         Q  R         R  S         S  U         $U  HV         RV  vW         W  X         X  Y         Y  Z         [  &\         0\  R]         \]  ~^         ^  _         _  `         `  b         b  .c         8c  Xd         bd  e         e  f         f  g         g   i         
i  *j         4j  Tk         ^k  ~l         l  m         m  n         ȑ                        1         4  U         X  y         |                      ę             	           -         0  Q         T  u         x                                              )         ,  M         P  q         t                        ݪ                      %         (  I         L  m         p                        ٲ         ܲ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               H!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     7
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            8     -,-h     ,-m-     m--    -;5-    ;5-C-    C-#w-l    #w---    ---x    -9b-	    9b--
    --
    -m-|    m-ճ-    ճ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               7!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        ?  \          ~                                    ;  1          T  J          m  c	          	  |
          
                                   /         0  W         X           4  N         O  i         j                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |        V"  |#         }#  $          *  +          +  ,          -  .          4.  D/          f/  v0         H  I         I  J         J  K         K  M         M  FN         GN  nO         oO  P         P  Q         Q  R         R  T         T  6U         7U  ^V         _V  W         W  X         X  Y         Y  Z         Z  &\         '\  N]         O]  v^         w^  _         _  `         `  a         a  c         c  >d         ?d  fe         ge  f         f  g         g  h         h  j         j  .k         /k  Vl         Wl  ~m         m  n         n  o         o  p         p  r         r  Fs         Gs  nt         ot  u         u  v         v  w         w  y         y  6z         7z  ^{         _{  |         |  }         }  ~         ~             &         '  N         O  v         w             ƅ         ǅ                        >         ?  f         g                        ލ         ߍ             -         .  T         U  |         }             ̔         ͔                        D         E  l         m                                              4         5  \         ]                        ԣ         գ             $         %  L         M  t         u             Ī         Ū                        <                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     p            !          H!  1          1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    b
         c
             5      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    e
         f
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            0     -,-X     ,-g-     g--    -)5-x    )5-C-|    C-w-\    w-a-    a--h    -?b-    ?b-m-	    m--h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               3!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        /  4          ^  c                                                          A  E	          l	  p
          
                                              -         .  H            :         ;  U         V  p         q                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                     %  C         J  h         o                                              !	         (	  F
         M
  k         r                                              $         +  I         P  n         u                                            	  '         .  L         S  q         x                                                *!         1!  O"          @G  @H          jH  jI         I  J         J  K         K  L         M  !N         (N  FO         MO  kP         rP  Q         Q  R         R  S         S  T         U  $V         +V  IW         PW  nX         uX  Y         Y  Z         Z  [         [  ]         	]  '^         .^  L_         S_  q`         x`  a         a  b         b  c         c  	e         e  /f         6f  Ug         \g  {h         h  i         i  j         j  k         k  m         m  ;n         Dn  co         lo  p         p  q         q  r         r  t         t  +u         4u  Sv         \v  {w         w  x         x  y         y  z         z  |         $|  C}         L}  k~         t~                      Ā                        3         <  [         d                        ӈ         ܈             #         ,  K         T  s         |                        Ώ  ΐ                               >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     l            !          L!  1          1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     8
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     --    -<-x    <-J-|    J-~-\    ~-.-    .--h    -Wb-    Wb-n-	    n--h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               >!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        5         6  P         Q  k         l                                       %          O  T          ~                          
          +  I          j                                               !   G!         H!  n"         o"  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            (         *  R         T  |         ~                                   $         &  N	         P	  x
         z
                                             "  J         L  t         v                                              F         H  p         r                                              B         D  l         n               !          .  /          /  1          51  S2          x2  3          3  4         @G  iH         jH  I         I  J         J  K         K  "M         $M  LN         NN  vO         xO  P         P  Q         Q  R         R  T          T  HU         JU  rV         tV  W         W  X         X  Y         Y  [         [  D\         F\  n]         p]  ^         ^  _         _  `         `  b         b  @c         Bc  jd         ld  e         e  f         f  g         g  i         i  <j         >j  fk         hk  l         l  m           ͏         Ώ             !         "  K         L  u         v             ɖ         ʖ                        G         H  q         r             ŝ         Ɲ                        C         D  m         n                      ¤                        ?         @  i         j                                              ;                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     s            !          V!  1          2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     O
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-m-     m--    -)5-x    )5-C-|    C-w-\    w-1-    1--h    -9b-    9b-m-	    m--h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               5!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        /  5          ^  d                                                          A  D	          l	  o
          
                                              -         .  H            :         ;  U         V  p         q                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                     %  D         J  i         o                                              "	         (	  G
         M
  l         r                                               %         +  J         P  o         u                                            	  (         .  M         S  r         x                                                +!         1!  P"         I  J         J  K         K  L         M  "N         (N  GO         MO  lP         rP  Q         Q  R         R  S         S   U         U  %V         +V  JW         PW  oX         uX  Y         Y  Z         Z  [         [  ]         	]  (^         .^  M_         S_  r`         x`  a         a  b         b  c         c  
e         e  0f         6f  Vg         \g  |h         h  i         i  j         j  k         k  m         m  :n         Dn  bo         lo  p         p  q         q  r         r  t         t  *u         4u  Rv         \v  zw         w  x         x  y         y  z         z  |         $|  B}         L}  j~         t~                      Ā             
           2         <  Z         d                        ҈         ܈             "         ,  J         T  r         |            "  "          L  L          v  v                      ʖ  ʗ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     k            !          H!  1          1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     7
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     --    -<-x    <-J-|    J-~-\    ~-.-    .--h    -Wb-    Wb-n-	    n--h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               >!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        5         6  P         Q  k         l                                       &          O  U          ~                          	          +  H          j                                    "          #   J!         K!  r"         s"  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            )         *  S         T  }         ~                                   %         &  O	         P	  y
         z
                                   !         "  K         L  u         v                                              G         H  q         r                                              C         D  m         n               !          4  6          A6  _7          7  8          8  9          
:  (;         @G  hH         jH  I         I  J         J  K         K  #M         $M  MN         NN  wO         xO  P         P  Q         Q  R         R  T          T  IU         JU  sV         tV  W         W  X         X  Y         Y  [         [  E\         F\  o]         p]  ^         ^  _         _  `         `  b         b  Ac         Bc  kd         ld  e         e  f         f  g         g  i         i  =j         >j  gk         hk  l         l  m           ̏         Ώ                       "  J         L  t         v             Ȗ         ʖ                        F         H  p         r             ĝ         Ɲ                        B         D  l         n                      ¤                        >         @  h         j                                              :                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     r            !          V!  1          2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     N
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     --    -G;-x    G;-I-|    I-/}-\    /}--    --h    -'b-    'b-x-	    x--h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ;!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         ;  U         V  p         q  	         	  
         
                                                         3          4            M  0          f  I            b            {                                  /  C           n   !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                        <         @  \         `  |                                                      	  
          
  <         @  \         `  |                                                                    <         @  \         `  |                                                                    <         @  \         `  |                        !         !  "         "  $         $  $%         +%  G&         N&  j'         q'  (         (  )         )  *         *  +         +  -          -  <.         C.  _/         f/  0         0  1         1  2         2  3         3  5         5  16         86  T7         [7  w8         ~8  9         9  :         :  ;         ;  =         
=  &>         ->  I?         P?  l@         s@  A         A  B         B  C         "D  >E         EE  aF         hF  G         G  H         I  J         K  L         &L  BM         IM  eN         lN  O         O  P         P  Q         Q  R         R  T         T  7U         >U  ZV         aV  }W         W  X         X  Y         Y  Z         Z  	\         \  ,]         3]  O^         V^  r_         y_  `         `  a         a  b         b  c         d  !e         (e  Df         Kf  gg          Lw  Ox          ~x  y          y  z          z  {          |  }                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     d            !          >!  1          1  q=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 u
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-;5-x    ;5-C-|    C-#w-\    #w-+-    +--h    -{b-    {b-n-	    n--h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               @!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        ?  ]          ~                                    ;  2          T  K          m  d	          	  }
          
                                   ,         -  S         T  z         *  D         E  _         `  z         {                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            *         ,  V         X                                              2         4  ^	         `	  
         
                                   :         <  f         h                                              B         D  n         p                                               J         L  v         x                          !         !  &#         HI  rJ         tJ  K         K  L         L  M         M  "O         $O  NP         PP  zQ         |Q  R         R  S         S  T          U  *V         ,V  VW         XW  X         X  Y         Y  Z         Z  \         \  2]         4]  ^^         `^  _         _  `         `  a         a  c         c  :d         <d  fe         he  f         f  g         g  h         h  j         j  Bk         Dk  nl         pl  m         m  n         n  o         o  q          q  Jr         Lr  vs         xs  t         t  u         u  v         v  &x         (x  Ry         Ty  ~z         z  {         {  |         |  ~         ~  .         0  Z         \                        ރ           
           6         8  b         d                                              >         @  j         l              3          X  s                      ة              3                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               D!         T!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     O
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-y-     y--    -S;-x    S;-I-|    I-;}-\    ;}--    --h    --b-    -b-x-	    x--h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               9!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         ;  U         V  p         q  	         	  
         
                                                          =          >             W  9          p  R            k                                              9  L           x   !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                        =         @  ]         `  }                                                      	  
          
  =         @  ]         `  }                                                                    =         @  ]         `  }                                                                    =         @  ]         `  }                        !         !  "         "  $         $  %%         +%  H&         N&  k'         q'  (         (  )         )  *         *  +         +  -          -  =.         C.  `/         f/  0         0  1         1  2         2  3         3  5         5  26         86  U7         [7  x8         ~8  9         9  :         :  ;         ;  =         
=  '>         ->  J?         P?  m@         s@  A         A  B         B  C         "D  ?E         EE  bF         hF  G         G  H         I  J         K   L         &L  CM         IM  fN         lN  O         O  P         P  Q         Q  R         R  T         T  8U         >U  [V         aV  ~W         W  X         X  Y         Y  Z         Z  
\         \  -]         3]  P^         V^  s_         y_  `         `  a         a  b         b  c         d  "e         (e  Ef         Kf  hg          F}  I~          x~  {                      ܀  ߁                               >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     e            !          B!  1          1  x=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 u
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-A5-x    A5-C-|    C-)w-\    )w-1-    1--h    -ub-    ub-n-	    n--h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               @!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        ?  \          ~                                    ;  1          T  J          m  c	          	  |
          
                                   /         0  W         X           4  N         O  i         j                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            +         ,  W         X                                              3         4  _	         `	  
         
                                   ;         <  g         h                                              C         D  o         p                                               K         L  w         x                          !         !  '#         HI  sJ         tJ  K         K  L         L  M         M  #O         $O  OP         PP  {Q         |Q  R         R  S         S  T          U  +V         ,V  WW         XW  X         X  Y         Y  Z         Z  \         \  3]         4]  _^         `^  _         _  `         `  a         a  c         c  ;d         <d  ge         he  f         f  g         g  h         h  j         j  Ck         Dk  ol         pl  m         m  n         n  o         o  q          q  Kr         Lr  ws         xs  t         t  u         u  v         v  'x         (x  Sy         Ty  z         z  {         {  |         |  ~         ~  /         0  [         \                        ߃                      7         8  c         d                                              ?         @  k         l            X  s                      خ              3          X  s                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               @!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     N
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-g-     g--    -55-x    55-C-|    C-w-\    w-
-    
--h    -9b-    9b-x-	    x--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               /!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        /  4          ^  c                                                          A  E	          l	  p
          
                                              -         .  H            :         ;  U         V  p         q                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         p  t                                                 $         e$  %         %  &         &  '         '  (         (  *         
*  $+         ++  E,         L,  f-         m-  .         .  /         /  0         0  1         1  3         3  ,4         34  M5         T5  n6         u6  7         7  8         8  9         9  :         :  <         <  4=         ;=  U>         \>  v?         }?  @         @  A         A  B         B  C         D  E         "E  <F         F  G         G  I         
I  $J         +J  EK         LK  fL         mL  M         M  N         N  O         O  P         P  R         R  ,S         3S  MT         TT  nU         uU  V         V  W         W  X         X  Y         Y  [         [  4\         ;\  U]         \]  v^         }^  _         _  `         `  a         a  b         c  d         "d  <e         Ce  ]f         df  ~g         g  h         h  j         "j  =k         Fk  al         jl  m         m  n         n  o         o  p         p  r         r  9s         Bs  ]t         ft  u         u  v         v  w         w  x         x  z         z  5{         >{  Y|         b|  }}         }  ~         ~                                   1         :  U         ^  y                               ʇ             	           -                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               !!         0!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 u
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-:-x    :-I-|    I-|-\    |-+-    +--h    -Eb-    Eb-	-	    	--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               H!         X!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                !  ;	         <	  V
         W
  q         I  c         d  ~                                                                   &             Q  K          |  v                                    
           4   9!          c!  h"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       7         8  S         T  o         p                                              	         	             3         4  O         P  k         l                                                                    /         0  K         L  g         h                                                                    +          ,   G!         !  "         "  #         #  %         %  :&         <&  Y'         ['  x(         z(  )         )  *         *  +         +  ,         ,  .         .  2/         4/  Q0         S0  p1         r1  2         2  3         3  4         4  5         5  7         7  *8         ,8  I9         K9  h:         j:  ;         ;  <         <  =         =  >         >  @         @  "A         $A  AB         CB  `C         C  E         E  4F         6F  SG         UG  rH         tH  I          l  m          m  n           o  $p          Np  Rq          |q  r         (  E         G  d         f                                 Õ                                    =         ?  \         ^  {         }                        ؟         ڟ                        5         7  T         V  s         u                        Ш         Ҩ                        -                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               %!         6!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   ;      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -? -X     ? -հ-     հ-ܰ-    ܰ-w/-p    w/-5;-T    5;-c-4    c-
-    
--8    -Eb-    Eb-x-	    x--8
    --׳-
                   =         b      J      7   U   7   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               -!         >!  01                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 64 })  outputElements  5184    input0Elements  31104   input1Elements  24576   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     4                           #         b      J      7   M     operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       @                        /  5          ^  d                                                          =  @	         h	  
         
                                                                                                  b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     4                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  1728    input0Elements  1728    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       x         L  P          x  |                          !         e$  %         %  &         &  '         '  (         (  *         
*  %+         ++  F,         L,  g-         m-  .         .  /         /  0         0  1         1  3         3  -4         34  N5         T5  o6         u6  7         7  8         8  9         9  :         :  <         <  5=         ;=  V>         \>  w?         }?  @         @  A         A  B         B  C         D  E         "E  =F         F  G         G  I         
I  %J         +J  FK         LK  gL         mL  M         M  N         N  O         O  P         P  R         R  -S         3S  NT         TT  oU         uU  V         V  W         W  X         X  Y         Y  [         [  5\         ;\  V]         \]  w^         }^  _         _  `         `  a         a  b         c  d         "d  =e         Ce  ^f         df  g         g  h         h  j         "j  <k         Fk  `l         jl  m         m  n         n  o         o  p         p  r         r  8s         Bs  \t         ft  u         u  v         v  w         w  x         x  z         z  4{         >{  X|         b|  |}         }  ~         ~                                   0         :  T         ^  x                               ʇ                        ,                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               %!         4!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 u
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-:-x    :-I-|    I-|-\    |-+-    +--h    -Kb-    Kb-	-	    	--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               D!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                !  ;	         <	  V
         W
  q         I  c         d  ~                                                                   &            Q  J          |  u                                               4   :!          c!  i"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       6         8  R         T  n         p                                              	         	             2         4  N         P  j         l                                                                    .         0  J         L  f         h                                                                    *          ,   F!         !  "         "  #         #  %         %  ;&         <&  Z'         ['  y(         z(  )         )  *         *  +         +  ,         ,  .         .  3/         4/  R0         S0  q1         r1  2         2  3         3  4         4  5         5  7         7  +8         ,8  J9         K9  i:         j:  ;         ;  <         <  =         =  >         >  @         @  #A         $A  BB         CB  aC         C  E         E  5F         6F  TG         UG  sH         tH  I          r  s          s  t          u  
v          4v  8w          bw  fx         (  F         G  e         f                                 Õ                                     >         ?  ]         ^  |         }                        ٟ         ڟ                        6         7  U         V  t         u                        Ѩ         Ҩ                        .                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               '!         :!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   <      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-a-     a--    -S;-x    S;-I-|    I-;}-\    ;}-e
-    e
--P    -Wb-    Wb-c-	    c-s-D
    s-3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ,!         <!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         ;  U         V  p         q  	         	  
         
                                                         3          4            M  0          f  I            b            {                                  /  C           n   !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            *         ,  V         X                                              2         4  ^	         `	  
         
                                   :         <  f         h                                              B         D  n         p                                               J         L  v         x                          !         !  &#         (#  R$         T$  ~%         %  &         &  '         '  )         )  .*         0*  Z+         \+  ,         ,  -         -  .         .  
0         0  61         81  b2         d2  3         3  4         4  5         5  7         7  >8         @8  j9         l9  :         :  ;         ;  <         <  >         >  F?         H?  r@         t@  A         A  B         B  C          F  G          G  H          $I  &J          RJ  TK          K  L         h  (j         *j  Tk         Vk  l         l  m         m  n         n  p         p  0q         2q  \r         ^r  s         s  t         t  u         u  w         w  8x         :x  dy         fy  z         z  {         {  |         |  ~         ~  @         B  l         n                      Ă                        C         E  o         q             ǉ         ɉ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               (!         @!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     --    -%<-x    %<-J-|    J-~-\    ~--    --h    -9b-    9b--	    --h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               <!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        5         6  P         Q  k         l                                       %          O  T          ~                          
          +  I          j                                               !   G!         H!  n"         o"  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       :         <  X         Z  v         x                                              
         
  *         ,  H         J  f         h                                                                    8         :  V         X  t         v                                              
           (         *  F         H  d          f   !         #  $         $  &         &  :'         <'  Z(         \(  z)         |)  *         *  +         +  ,         ,  -         -  /         /  :0         <0  Z1         \1  z2         |2  3         3  4         4  5         5  6         6  8         8  :9         <9  Z:         \:  z;         |;  <         <  =         =  >         >  ?         ?  A         A  :B         <B  ZC         \C  zD         |D  E         F  G         G  H         H  J         J  ?K         @K  eL         fL  M         M  N         N  O         O  P         P  #R         $R  ES         GS  hT         jT  U         U  V         V  W         W  X         X  Z         Z  :[         <[  ]\         _\  ]         ]  ^         ^  _         _  `         `  b         b  %c         'c  Fd         Hd  ge         ie  f         f  g         g  h          2v  Fw          pw  x          x  y          y   {          *{  >|                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                                !         *!  1         1  z=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   ;      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-a-     a--    -Y;-x    Y;-I-|    I-A}-\    A}-k
-    k
--P    -Wb-    Wb-c-	    c-s-D
    s-3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               +!         :!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         ;  U         V  p         q  	         	  
         
                                                          =          >             W  9          p  R            k                                              9  L           x   !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            +         ,  W         X                                              3         4  _	         `	  
         
                                   ;         <  g         h                                              C         D  o         p                                               K         L  w         x                          !         !  '#         (#  S$         T$  %         %  &         &  '         '  )         )  /*         0*  [+         \+  ,         ,  -         -  .         .  0         0  71         81  c2         d2  3         3  4         4  5         5  7         7  ?8         @8  k9         l9  :         :  ;         ;  <         <  >         >  G?         H?  s@         t@  A         A  B         B  C          L  M          M  N          
O  P          8P  :Q          fQ  hR         h  )j         *j  Uk         Vk  l         l  m         m  n         n  p         p  1q         2q  ]r         ^r  s         s  t         t  u         u  w         w  9x         :x  ey         fy  z         z  {         {  |         |  ~         ~  A         B  m         n             Â         Ă                        D         E  p         q             ȉ         ɉ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               *!         @!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-y-     y--    -+<-x    +<-J-|    J-~-\    ~--    --h    -?b-    ?b--	    --h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               8!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        5         6  P         Q  k         l                                       &          O  U          ~                          	          +  H          j                                    "          #   J!         K!  r"         s"  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       ;         <  Y         Z  w         x                                              
         
  +         ,  I         J  g         h                                                                    9         :  W         X  u         v                                                         )         *  G         H  e          f   !         #  $         $  &         &  ;'         <'  [(         \(  {)         |)  *         *  +         +  ,         ,  -         -  /         /  ;0         <0  [1         \1  {2         |2  3         3  4         4  5         5  6         6  8         8  ;9         <9  [:         \:  {;         |;  <         <  =         =  >         >  ?         ?  A         A  ;B         <B  [C         \C  {D         |D  E         F  G         G  H         H  J         J  >K         @K  dL         fL  M         M  N         N  O         O  P         P  "R         $R  FS         GS  iT         jT  U         U  V         V  W         W  X         X  Z         Z  ;[         <[  ^\         _\  ]         ]  ^         ^  _         _  `         `  b         b  &c         'c  Gd         Hd  he         ie  f         f  g         g  h          h|  |}          }  ~          ~            "  6          `  t                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               "!         .!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   <      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,-s-     s--    -;-    ;-J-    J-}-d    }-A-    A--p    -ub-    ub-	-
    	--
    --t    -3׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               5!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        9         :  V         W  s         t            |                                    9  M          x                                                D  I          s  x                                   -!         .!  H"         I"  c#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         ,            J  <          h  Z            x                     %  1&         2&  ['         \'  (         (  )         )  *         *  ,         ,  --         .-  W.         X.  /         /  0         0  1         1  2          3  )4         *4  S5         T5  }6         ~6  7         7  8         8  9         9  %;         &;  O<         P<  y=         z=  >         >  ?         ?  @         @  !B         "B  KC         LC  uD         vD  E         E  F         F  G         G  I         I  FJ         HJ  pK         rK  L         L  M         M  N         N  P         P  BQ         DQ  lR         nR  S         S  T         T  U         U  W         W  >X         @X  hY         jY  Z         Z  [         [  \         \  ^         ^  :_         <_  d`         f`  a         a  b         b  c         c  e         e  6f         8f  `g         bg  h         h  i         i  j         j  l         
l  2m         4m  \n         ^n  o         o  p         p  q         q  s         s  .t         0t  Xu         Zu  v         v  w         w  x         x   z         z  *{         ,{  T|         V|  ~}         }  ~         ~                        &         (  P         R  z         |             Ά         І             "         $  L         N  v         x             ʍ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               5!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    b
         c
             /      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     H
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-y-     y--    -:-x    :-I-|    I-|-\    |-[-    [--P    -ob-    ob-c-	    c-a-D
    a-3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               9!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                !  ;	         <	  V
         W
  q         I  c         d  ~                                                                   &             Q  K          |  v                                    
           4   9!          c!  h"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p                        ?         @  _         `                                                        	  
          
  ?         @  _         `                                                                      ?         @  _         `                                                                      ?         @  _         `                          !         !  "         "  #          $  %          %  >&         @&  ^'         `'  ~(         (  )         )  *         *  +         +  ,          -  .          .  >/         @/  ^0         `0  ~1         1  2         2  3         3  4         4  5          6  7          7  >8         @8  ^9         `9  ~:         :  ;         ;  <         <  =         =  >          ?  @          @  >A         G  
I         I  *J         +J  JK         KK  jL         kL  M         M  N         N  O         O  P         P  
R         R  *S         +S  JT         KT  jU         kU  V         V  W         W  X         X  Y         Y  
[         [  *\         +\  J]         K]  j^         k^  _         _  `         `  a         a  b         b  d         d  6e         7e  Xf         Yf  zg         {g  h            ׆                      D  ]                      ʊ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               5!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 y
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     --    -;-    ;-J-    J-}-d    }-A-    A--p    -ub-    ub-	-
    	--
    --t    -3׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               9!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        ;         <  Y         Z  w         x                                                >  Q          }                                                 I  O          x  ~                                   2!         3!  M"         N"  h#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                                         :  ,         %  0&         2&  Z'         \'  (         (  )         )  *         *  ,         ,  ,-         .-  V.         X.  /         /  0         0  1         1  2          3  (4         *4  R5         T5  |6         ~6  7         7  8         8  9         9  $;         &;  N<         P<  x=         z=  >         >  ?         ?  @         @   B         "B  JC         LC  tD         vD  E         E  F         F  G         G  I         I  GJ         HJ  qK         rK  L         L  M         M  N         N  P         P  CQ         DQ  mR         nR  S         S  T         T  U         U  W         W  ?X         @X  iY         jY  Z         Z  [         [  \         \  ^         ^  ;_         <_  e`         f`  a         a  b         b  c         c  e         e  7f         8f  ag         bg  h         h  i         i  j         j  	l         
l  3m         4m  ]n         ^n  o         o  p         p  q         q  s         s  /t         0t  Yu         Zu  v         v  w         w  x         x  z         z  +{         ,{  U|         V|  }         }  ~         ~                        '         (  Q         R  {         |             φ         І             #         $  M         N  w         x             ˍ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               5!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    b
         c
             0      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     I
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-y-     y--    -:-x    :-I-|    I-|-\    |--6-    -6--P    -ob-    ob-c-	    c-a-D
    a-3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               6!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                !  ;	         <	  V
         W
  q         I  c         d  ~                                                                   &            Q  J          |  u                                               4   :!          c!  i"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p                        >         @  ^         `  ~                                                      	  
          
  >         @  ^         `  ~                                                                    >         @  ^         `  ~                                                                    >         @  ^         `  ~                        !         !  "         "  #          $  %          %  ?&         @&  _'         `'  (         (  )         )  *         *  +         +  ,          -  .          .  ?/         @/  _0         `0  1         1  2         2  3         3  4         4  5          6  7          7  ?8         @8  _9         `9  :         :  ;         ;  <         <  =         =  >          ?  @          @  ?A         G  	I         I  )J         +J  IK         KK  iL         kL  M         M  N         N  O         O  P         P  	R         R  )S         +S  IT         KT  iU         kU  V         V  W         W  X         X  Y         Y  	[         [  )\         +\  I]         K]  i^         k^  _         _  `         `  a         a  b         b  d         d  5e         7e  Wf         Yf  yg         {g  h            &          ұ              .          X  q                               >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               5!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 y
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    <     -,-p     ,--     --    -_5-    _5-C-    C-Gw-t    Gw-12-     12-9-X    9--    -ub-p	    ub-	-x
    	--
    -s-    s-ճ-\    ճ--<                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               7!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        /  4          ^  c                                                          A  B	          l	  m
          
                                              -         .  H            A         B  c         d                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        $  %         K  L         L  N         N  EO         FO  mP         nP  Q         Q  R         R  S         S  U         U  5V         6V  ]W         ^W  X         X  Y         Y  Z         Z  [         [  %]         &]  M^         N^  u_         v_  `         `  a         a  b         b  d         d  =e         >e  ef         ff  g         g  h         h  i         i  k         k  -l         .l  Um         Vm  }n         ~n  o         o  p         p  q         q  s         s  Et         Ft  mu         nu  v         v  w         w  x         x  z         z  5{         6{  ]|         ^|  }         }  ~         ~                        %         &  M         N  u         v             ņ         Ɔ                        =         >  e         f                        ݎ         ގ             -         .  T         V  |         ~             ̕         Ε                        D         F  l         n                                              4         6  \         ^                        Ԥ         ֤             $          &  '          N  O          v  w                      ƫ  Ǭ                      <         >  d         f                                           b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               8!         H!  1         1  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    b
         c
             /                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    b
         c
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     l
            8     -,-h     ,-y-     y--    -<-    <-K-    K-m~-l    m~--    -M-P    M--    -b-h	    b-!-p
    !-/-
    /-ᰳ-    ᰳ-9׳-L                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               6!         >!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        5         6  P         Q  k         l                                       %          O  T          ~                                    +  D          j                         '         (  Y          Z   !         !  "         "  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h         $  %         :%  [&         \&  }'         ~'  (         (  )         )  *         *  ,         ,  '-         (-  I.         J.  k/         l/  0         0  1         1  2         2  3         3  5         5  76         86  Y7         Z7  {8         |8  9         9  :         :  ;         ;  =         =  %>         &>  G?         H?  i@         j@  A         A  B         B  C         C  D         D  F         F  5G          6G  1H          XH  SI          zI  uJ          J  K         K  L         L  "N         $N  UO         WO  P         P  Q         Q  R         R  !T         #T  TU         VU  V         V  W         W  X         X  'Z         (Z  ][         ^[  \         \  ]         ]  ^          _  5`         6`  ka         la  b         b  c         c  e         e  ?f         @f  sg         tg  h         h  i         i  k         k  Cl         Dl  wm         xm  n         n  o         o   q         q  "r         $r  Ds         Fs  ft         ht  u         u  v         v  w         w  x         x  z         z  2{         4{  T|         V|  v}         x}  ~         ~             ܀         ހ                        "  B         D  d         f                        ʈ         ̈                        0                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               B!         V!  1         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    b
         c
             D                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     ;
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    <     -,-p     ,--     -#-    #-_5-    _5-C-    C-Gw-t    Gw-2-     2-9-X    9--    -ub-p	    ub-	-x
    	--
    -s-    s-ճ-\    ճ--<                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ;!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        /  5          ^  d                                                          A  C	          l	  n
          
                                              -         .  H            @         A  a         b                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        $  %         K  L         L  N         N  DO         FO  lP         nP  Q         Q  R         R  S         S  U         U  4V         6V  \W         ^W  X         X  Y         Y  Z         Z  [         [  $]         &]  L^         N^  t_         v_  `         `  a         a  b         b  d         d  <e         >e  df         ff  g         g  h         h  i         i  k         k  ,l         .l  Tm         Vm  |n         ~n  o         o  p         p  q         q  s         s  Dt         Ft  lu         nu  v         v  w         w  x         x  z         z  4{         6{  \|         ^|  }         }  ~         ~                        $         &  L         N  t         v             Ć         Ɔ                        <         >  d         f                        ܎         ގ             ,         .  U         V  }         ~             ͕         Ε                        E         F  m         n                                              5         6  ]         ^                        դ         ֤             %         &  M         N  u         v             ū         ƫ                                    >  ?          f  g                                          b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               :!         H!  1         1  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    b
         c
             0                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    b
         c
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     k
            8     -,-h     ,-s-     s--    -<-    <-K-    K-m~-l    m~--    -M-P    M--    -b-h	    b-!-p
    !-/-
    /-ᰳ-    ᰳ-9׳-L                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               4!         >!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        5         6  P         Q  k         l                                       &          O  U          ~                                    +  C          j                         #         $  V          W   !         !  "         "  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        $  9%         :%  Z&         \&  |'         ~'  (         (  )         )  *         *  ,         ,  &-         (-  H.         J.  j/         l/  0         0  1         1  2         2  3         3  5         5  66         86  X7         Z7  z8         |8  9         9  :         :  ;         ;  =         =  $>         &>  F?         H?  h@         j@  A         A  B         B  C         C  D         D  F         F  4G         6G  WH         XH  yI         zI  J         J  K         K  L         L  #N         $N  VO         WO  P         P  Q         Q  R         R  "T         #T  UU         VU  V         V  W         W  X         X  &Z         (Z  \[         ^[  \         \  ]         ]  ^          _  4`         6`  ja         la  b         b  c         c  
e         e  >f         @f  rg         tg  h         h  i         i  k         k  Bl         Dl  vm         xm  n         n  o          o  p          q  q          $r  s          Fs  At          ht  cu         u  v         v  w         w  x         x  z         z  3{         4{  U|         V|  w}         x}  ~         ~             ݀         ހ              !         "  C         D  e         f                        ˈ         ̈                        1                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               @!         V!  1         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    b
         c
             F                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     <
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-<-x    <-J-|    J-~-\    ~-4-    4--h    -b-    b-n-	    n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               G!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             !         "  C         D  e         f                                                           R  [                                              .  3          ]  b                                  +          ,   F!         G!  a"         b"  |#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            $         &  J         L  p         r                                            
  .	         0	  T
         V
  z         |                                              8         :  ^         `                                                         B         D  h         j                                               &          (   L!         N!  r"          %  &          '  6(          `(  y)          )  *          *  +         K  L         L  M          N  %O         &O  KP         LP  qQ         rQ  R         R  S         S  T         T  	V         
V  /W         0W  UX         VX  {Y         |Y  Z         Z  [         [  \         \  ^         ^  9_         :_  _`         ``  a         a  b         b  c         c  d         d  f         f  Cg         Dg  ih         jh  i         i  j         j  k         k  m         m  'n         (n  Mo           ?         @  e         f                        י         ؙ             #         $  I         J  o         p                                              -         .  S         T  y         z             Ũ         ƨ                        7         8  ]         ^                        ϰ         а                        A         B  g                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               H!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     -
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     -)-    )-S5-    S5-C-    C-;w-l    ;w-5-    5--x    -b-	    b--
    --
    -y-|    y-ճ-    ճ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               A!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        ?  U          ~                                    ;  "          T  ;          m  T	          	  m
          
                                              3         4  R         *  D         E  _         `  z         {                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         h  b                                                         8"  Y#         [#  |$         ~$  %         %  &         &  '         '  )         )  0*         2*  T+         V+  x,         z,  -         -  .         .  /         /  1         
1  ,2         .2  P3         R3  t4         v4  5         5  6         6  7         7  9         9  (:         *:  L;         N;  p<         r<  =         =  >         >  ?         ?   A         A  $B         &B  HC         JC  lD         nD  E         E  F         F  G         -l  Nm         Pm  qn         sn  o         o  p         p  q         q  s         s  %t         't  Iu         Ku  mv         ov  w         w  x         x  y         y  z         z  !|         #|  E}         G}  i~         k~                        Ձ         ׁ                        A         C  e         g                        щ         Ӊ           ړ                       "  D         F  h         j                        ԛ         ֛                        @         B  d         f                        У         ң                        <         >  `         b                        ̫         Ϋ                        8         :  \         ^                        ȳ         ʳ                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               @!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    b
         c
             ,      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    b
         c
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     [
            0     -,-X     ,--     -5-    5-<-x    <-J-|    J-~-\    ~-4-    4--h    -b-    b-n-	    n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               G!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             "         #  E         F  h         i                                                            W  _                                    
          3  9          b  h                                  0          1   K!         L!  f"         g"  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            %         &  K         L  q         r                                   	         
  /	         0	  U
         V
  {         |                                              9         :  _         `                                                         C         D  i         j                                              '          (   M!         N!  s"          ),  B-          l-  .          .  /          /  1          51  N2         K  L         L  M          N  $O         &O  JP         LP  pQ         rQ  R         R  S         S  T         T  V         
V  .W         0W  TX         VX  zY         |Y  Z         Z  [         [  \         \  ^         ^  8_         :_  ^`         ``  a         a  b         b  c         c  d         d  f         f  Bg         Dg  hh         jh  i         i  j         j  k         k   m         m  &n         (n  Lo           >         @  d         f                        ֙         ؙ             "         $  H         J  n         p                                              ,         .  R         T  x         z             Ĩ         ƨ                        6         8  \         ^                        ΰ         а                        @         B  f                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               J!         R!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     ,
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     -)-    )-Y5-    Y5-C-    C-Aw-l    Aw-5-    5--x    -{b-	    {b--
    --
    -y-|    y-ճ-    ճ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               A!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        ?  T          ~                                    ;  !          T  :          m  S	          	  l
          
                                              7         8  W         4  N         O  i         j                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                     @  :          d  ^                                 8"  Z#         [#  }$         ~$  %         %  &         &  '         '  )         )  1*         2*  U+         V+  y,         z,  -         -  .         .  /         /  	1         
1  -2         .2  Q3         R3  u4         v4  5         5  6         6  7         7  9         9  ):         *:  M;         N;  q<         r<  =         =  >         >  ?         ?  A         A  %B         &B  IC         JC  mD         nD  E         E  F         F  G         -l  Om         Pm  rn         sn  o         o  p         p  q         q  s         s  &t         't  Ju         Ku  nv         ov  w         w  x         x  y         y  z         z  "|         #|  F}         G}  j~         k~                        ց         ׁ                        B         C  f         g                        ҉         Ӊ           ړ             !         "  E         F  i         j                        ՛         ֛                        A         B  e         f                        ѣ         ң                        =         >  a         b                        ͫ         Ϋ                        9         :  ]         ^                        ɳ         ʳ                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               >!         @!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    b
         c
             -      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    b
         c
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     \
            0     -,-X     ,--     -)-    )-A5-x    A5-C-|    C-)w-\    )w--    --h    -{b-    {b-!n-	    !n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               C!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        /  4          ^  c                                                          A  =	          l	  h
          
                                              -         .  H            :         ;  U         V  p         q                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            "         $  F         H  j         l                                              	          	  B
         D
  f         h                                                         >         @  b         d                                                         :         <  ^         `                                              !         !  6"          /  0          0  1          1  3          ,3  :4          d4  r5         .I  PJ         RJ  tK         vK  L         L  M         M  N         N  P         P  (Q         *Q  LR         NR  pS         rS  T         T  U         U  V         V   X         X  $Y         &Y  HZ         JZ  l[         n[  \         \  ]         ]  ^         ^  _         _  #a         $a  Ib         Jb  oc         pc  d         d  e         e  f         f  h         h  -i         .i  Sj         Tj  yk         m  n         n  p         p  <q         =q  `r         ar  s         s  t         t  u         u  v         v  x         x  8y         9y  \z         ]z  {         {  |         |  }         }  ~         ~             4         5  X         Y  |         }             ą         Ņ                        0         1  T         U  x         y                                                      >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ?!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     @
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-<-x    <-J-|    J-}-\    }--    --h    -b-    b--n-	    -n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               @!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        5         6  P         Q  k         l                                       %          O  T          ~                                    +  A          j                                                 /!         0!  N"         O"  m#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            $         &  J         L  p         r                                            
  .	         0	  T
         V
  z         |                                              8         :  ^         `                                                         B         D  h         j                                               &          (   L!         N!  r"         t"  #         #  $         $  %         %  '         '  1(         2(  W)         X)  }*         ~*  +         +  ,         ,  -         -  /         /  ;0         <0  a1         b1  2          J  K          K  L           M  
N          4N  >O          hO  rP         _j  k         k  l         l  m         m  n         n  p         p  Aq         Cq  gr         ir  s         s  t         t  u         u  v         w  %x         'x  Ky         My  qz         sz  {         {  |         |  }         }  	           /         1  U         W  {         }             Ǆ         Ʉ                        9         ;  _         a                        ь         ӌ                      	  .         /  T         U  z         {             ƕ         Ǖ                        8         9  ^         _                        Н         ѝ                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               A!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     O
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-A5-x    A5-C-|    C-)w-\    )w--    --h    -{b-    {b-!n-	    !n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               C!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        /  5          ^  d                                                          A  <	          l	  g
          
                                              -         .  H            :         ;  U         V  p         q                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            #         $  G         H  k         l                                              	          	  C
         D
  g         h                                                         ?         @  c         d                                                         ;         <  _         `                                              !         !  7"          5  6          6  7          8  9          D9  R:          |:  ;         .I  QJ         RJ  uK         vK  L         L  M         M  N         N  P         P  )Q         *Q  MR         NR  qS         rS  T         T  U         U  V         V  X         X  %Y         &Y  IZ         JZ  m[         n[  \         \  ]         ]  ^         ^  _         _  "a         $a  Hb         Jb  nc         pc  d         d  e         e  f         f  h         h  ,i         .i  Rj         Tj  xk         m  n         n  p         p  ;q         =q  _r         ar  s         s  t         t  u         u  v         v  x         x  7y         9y  [z         ]z  {         {  |         |  }         }  ~         ~             3         5  W         Y  {         }             Å         Ņ                        /         1  S         U  w         y                                                      >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ?!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     ?
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-<-x    <-J-|    J-}-\    }--    --h    -b-    b--n-	    -n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               B!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        5         6  P         Q  k         l                                       &          O  U          ~                                    +  @          j                                                 2!         3!  R"         S"  r#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            %         &  K         L  q         r                                   	         
  /	         0	  U
         V
  {         |                                              9         :  _         `                                                         C         D  i         j                                              '          (   M!         N!  s"         t"  #         #  $         $  %         %  
'         '  0(         2(  V)         X)  |*         ~*  +         +  ,         ,  -         -  /         /  :0         <0  `1         b1  2          P  Q          Q  R          S  T          8T  BU          lU  vV         _j  k         k  l         l  m         m  n         n  p         p  Bq         Cq  hr         ir  s         s  t         t  u         u   w         w  &x         'x  Ly         My  rz         sz  {         {  |         |  }         }  
           0         1  V         W  |         }             Ȅ         Ʉ                        :         ;  `         a                        Ҍ         ӌ                      	  -         /  S         U  y         {             ŕ         Ǖ                        7         9  ]         _                        ϝ         ѝ                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               A!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     N
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-_;-x    _;-I-|    I-G}-\    G}-+-    +--h    -ub-    ub-x-	    x--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               A!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         ;  U         V  p         q  	         	  
         
           {                                              3          4  '          M  @          f  Y            r                                              /  K           n   !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                      "  B         D  d         f                                              	         	  0
         2
  R         T  t         v                                                          @         B  b         d                                                         .         0  P         R  r         t                                     !         "  1#         3#  S$         sG  H         H  I         I  J         J  K         K  M         M  =N         ?N  _O         aO  P         P  Q         Q  R         R  S         S  	U         U  +V         -V  MW         OW  oX         qX  Y         Y  Z         Z  [         [  \         \  ^         ^  ;_         =_  ]`         _`  a         a  b         b  c         c  d         d  f         	f  )g         +g  Kh         Mh  mi         oi  j         j  k         k  l                      Ņ                          
          1  .         $  D         F  f         h                        ̕         Ε                        2         4  T         V  v         x                        ܞ         ޞ                        "  B         D  d         f                        ʦ         ̦                        0         2  R         T  t                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               :!         F!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 u
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-G5-x    G5-C-|    C-/w-\    /w-C-    C--h    -b-    b--n-	    -n--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               @!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        ?  U          ~                                    ;  "          T  ;          m  T	          	  m
          
                                              3         4  R         *  D         E  _         `  z         {                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            &         (  N         P  v         x                                              >	         @	  f
         h
                                              .         0  V         X  ~                                                        F         H  n         p                                              6         8  ^          `   !         !  "         "  $         $  :%         <%  b&         d&  '         '  (         (  )         )  +         +  *,         ,,  R-         T-  z.         |.  /         /  0         0  1         1  3         3  B4         D4  j5         l5  6         6  7         7  8         8  
:         :  2;         4;  Z<         \<  =         =  >         >  ?         ?  @         @  "B         $B  JC         LC  rD         tD  E         E  F          :H  II          vI  J          J  K          K  L          *M  9N         Bm  hn         jn  o         o  p         p  q         q  s         
s  0t         2t  Xu         Zu  v         v  w         w  x         x  y         y   {         "{  H|         J|  p}         r}  ~         ~                                   8         :  `         b                        ؇         ڇ              (         *  P         R  x         z             Ȏ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               @!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     O
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-k;-x    k;-J-|    J-S}-\    S}-+-    +--h    -{b-    {b-x-	    x--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               =!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         ;  U         V  p         q  	         	  
         
                                                         =          >  0          W  I          p  b            {                                              9  T           x   !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            !         "  C         D  e         f                                              	         	  1
         2
  S         T  u         v                                                          A         B  c         d                                                         /         0  Q         R  s         t                                     !         "  2#         3#  T$         sG  H         H  I         I  J         J  K         K  M         M  >N         ?N  `O         aO  P         P  Q         Q  R         R  S         S  
U         U  ,V         -V  NW         OW  pX         qX  Y         Y  Z         Z  [         [  \         \  ^         ^  <_         =_  ^`         _`  a         a  b         b  c         c  d         d  f         	f  *g         +g  Lh         Mh  ni         oi  j         j  k         k  l          U  R          y  v                                             $  E         F  g         h                        ͕         Ε                        3         4  U         V  w         x                        ݞ         ޞ              !         "  C         D  e         f                        ˦         ̦                        1         2  S         T  u                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               <!         J!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 u
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-S5-x    S5-C-|    C-;w-\    ;w-C-    C--h    -{b-    {b-'n-	    'n--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               B!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        ?  T          ~                                    ;  !          T  :          m  S	          	  l
          
                                              7         8  W         4  N         O  i         j                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            '         (  O         P  w         x                                              ?	         @	  g
         h
                                              /         0  W         X                                                          G         H  o         p                                              7         8  _          `   !         !  "         "  $         $  ;%         <%  c&         d&  '         '  (         (  )         )  +         +  +,         ,,  S-         T-  {.         |.  /         /  0         0  1         1  3         3  C4         D4  k5         l5  6         6  7         7  8         8  :         :  3;         4;  [<         \<  =         =  >         >  ?         ?  @         @  #B         $B  KC         LC  sD         tD  E         E  F          fN  uO          O  P          P  Q          R  )S          VS  eT         Bm  in         jn  o         o  p         p  q         q  	s         
s  1t         2t  Yu         Zu  v         v  w         w  x         x  y         y  !{         "{  I|         J|  q}         r}  ~         ~                                   9         :  a         b                        ه         ڇ             )         *  Q         R  y         z             Ɏ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               >!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     N
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-;5-x    ;5-C-|    C-#w-\    #w--    --h    -]b-    ]b-x-	    x--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               A!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        /  4          ^  c                                                          A  =	          l	  h
          
                                              -         .  H            :         ;  U         V  p         q                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                        0  0          `  `                                 "  $         $  *%         ,%  J&         L&  j'         l'  (         (  )         )  *         *  +         +  
-         -  *.         ,.  J/         L/  j0         l0  1         1  2         2  3         3  4         4  
6         6  *7         ,7  J8         L8  j9         l9  :         :  ;         ;  <         <  =         =  
?         ?  *@         ,@  JA         LA  jB         lB  C         C  D         D  E         AF  _G         aG  H         H  I         I  J         J  K         K  L         M  N         !N  ?O         AO  _P         aP  Q         Q  R         R  S         S  T         T  U         V  W         !W  ?X         AX  _Y         aY  Z         Z  [         [  \         \  ]         ]  ^         _  `         !`  ?a         Aa  _b         ab  c         c  d         d  e         e  f         f  g         h  i         i  k         k  1l         2l  Qm         Rm  qn         rn  o         o  p         p  q         q  r         r  t         t  1u         2u  Qv         Rv  qw         rw  x         x  y         y  z         z  {         {  }         }  1~         2~  Q         R  q         r                        у         ҃                        1         2  Q         R  q                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               1!         8!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 u
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-:-x    :-I-|    I-|-\    |-i-    i--h    -]b-    ]b-x-	    x--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               B!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                !  ;	         <	  V
         W
  q         I  c         d  ~                                                                   &  (          Q  S          |  ~                                    
           4   9!          c!  h"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            )         *  S         T  }         ~                                   %         &  O	         P	  y
         z
                                   !         "  K         L  u         v                                              G         H  q         r                                              C         D  m         n               !         !  "         $  <%         >%  f&         h&  '         '  (         (  )         )  +         +  8,         :,  b-         d-  .         .  /         /  0         0  
2         2  43         63  ^4         `4  5         5  6         6  7         7  9         9  0:         2:  Z;         \;  <         <  =         =  >         >  @         @  ,A         .A  VB         XB  C         C  D         D  E         E  F          G  (H         :H  bI         dI  J         J  K         K  L         L  
N         N  4O         6O  ^P         `P  Q         Q  R         R  S         S  U         U  0V         2V  ZW         \W  X         X  Y         Y  Z         Z  \         \  ,]         .]  V^         X^  _         _  `         `  a         a  b          c  (d         *d  Re         Te  |f         ~f  g         g  h          r  s          t  u          2u  (v          Rv  Hw          rw  hx         ʎ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               +!         >!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     --    -;5-x    ;5-C-|    C-#w-\    #w--    --h    -cb-    cb-x-	    x--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               =!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        /  5          ^  d                                                          A  <	          l	  g
          
                                              -         .  H            :         ;  U         V  p         q                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                   P  P	          	  
          
           "  
$         $  +%         ,%  K&         L&  k'         l'  (         (  )         )  *         *  +         +  -         -  +.         ,.  K/         L/  k0         l0  1         1  2         2  3         3  4         4  6         6  +7         ,7  K8         L8  k9         l9  :         :  ;         ;  <         <  =         =  ?         ?  +@         ,@  KA         LA  kB         lB  C         C  D         D  E         AF  `G         aG  H         H  I         I  J         J  K         K   M         M   N         !N  @O         AO  `P         aP  Q         Q  R         R  S         S  T         T   V         V   W         !W  @X         AX  `Y         aY  Z         Z  [         [  \         \  ]         ]   _         _   `         !`  @a         Aa  `b         ab  c         c  d         d  e         e  f         f   h         h  i         i  k         k  0l         2l  Pm         Rm  pn         rn  o         o  p         p  q         q  r         r  t         t  0u         2u  Pv         Rv  pw         rw  x         x  y         y  z         z  {         {  }         }  0~         2~  P         R  p         r                        Ѓ         ҃                        0         2  P         R  p                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               3!         <!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 u
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-:-x    :-I-|    I-|-\    |-i-    i--h    -cb-    cb-x-	    x--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               @!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                !  ;	         <	  V
         W
  q         I  c         d  ~                                                                   &  '          Q  R          |  }                                               4   :!          c!  i"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            (         *  R         T  |         ~                                   $         &  N	         P	  x
         z
                                             "  J         L  t         v                                              F         H  p         r                                              B         D  l         n               !         !  "         $  =%         >%  g&         h&  '         '  (         (  )         )  +         +  9,         :,  c-         d-  .         .  /         /  0         0  2         2  53         63  _4         `4  5         5  6         6  7         7  9         9  1:         2:  [;         \;  <         <  =         =  >         >  @         @  -A         .A  WB         XB  C         C  D         D  E         E  F          G  )H         :H  cI         dI  J         J  K         K  L         L  N         N  5O         6O  _P         `P  Q         Q  R         R  S         S  U         U  1V         2V  [W         \W  X         X  Y         Y  Z         Z  \         \  -]         .]  W^         X^  _         _  `         `  a         a  b          c  )d         *d  Se         Te  }f         ~f  g         g  h          x  y          y  z          z  {          {  |          }  ~         ʎ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               /!         B!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-[-     [--    -Y;-x    Y;-I-|    I-A}-\    A}--    --h    -]b-    ]b-E-	    E--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               *!         4!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         ;  U         V  p         q  	         	  
         
           {                                              3          4  '          M  @          f  Y            r                                              /  K           n   !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       :         <  X         Z  v         x                                              
         
  *         ,  H         J  f         h                                                                    8         :  V         X  t         v                                              
           (         *  F         H  d          f   !         p"  #         #  $         $  %         %  &         &  (         (  ")         $)  @*         B*  ^+         `+  |,         ~,  -         -  .         .  /         /  0         0  2         2  03         23  N4         P4  l5         n5  6         6  7         7  8         8  9         9  ;         ;   <         "<  >=         @=  \>         ^>  z?         |?  @         @  A         A  B         B  C         C  E         E  .F         0F  LG         :H  VI         XI  tJ         vJ  K         K  L         L  M         M  N         N  
P         P  (Q         *Q  FR         HR  dS         fS  T         T  U         U  V         V  W         W  X         X  Z         Z  6[         8[  T\         V\  r]         t]  ^         ^  _         _  `         `  a         a  c         
c  &d         (d  De         Fe  bf          y  z          z  {          |  *}          N}  f~          ~                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               .!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   r      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-s-     s--    -A5-x    A5-C-|    C-)w-\    )w--    --h    -?b-    ?b-x-	    x--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               4!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        ?  U          ~                                    ;  "          T  ;          m  T	          	  m
          
                                              3         4  R         *  D         E  _         `  z         {                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                                         0           &  '         '  (         (  
*         *  6+         7+  b,         c,  -         -  .         .  /         /  1         1  >2         ?2  j3         k3  4         4  5         5  6         6  8         8  F9         G9  r:         s:  ;         ;  <         <  =         =  !?         #?  M@         O@  yA         {A  B         B  C         C  D         D  )F         +F  UG         WG  H         H  I         I  J         J  L         L  1M         3M  ]N         _N  O         O  P         P  Q         Q  S         S  9T         ;T  eU         gU  V         V  W         W  X         X  Z         Z  A[         C[  m\         o\  ]         ]  ^         ^  _         _  a         a  Ib         Kb  uc         wc  d         d  e         e  f         f  %h         'h  Qi         Si  }j         j  k         k  l          n  *o         ,o  Vp         Xp  q         q  r         r  s         s  u         u  2v         4v  ^w         `w  x         x  y         y  z         z  |         |  :}         <}  f~         h~                                              B         D  n         p             Ƈ         ȇ                         J         L  v         x             Ύ         Ў             &                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               "!         0!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-U-     U--    -e;-x    e;-I-|    I-M}-\    M}-C-    C--h    -Wb-    Wb-E-	    E--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               '!         2!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         ;  U         V  p         q  	         	  
         
                                                         =          >  0          W  I          p  b            {                                              9  T           x   !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       ;         <  Y         Z  w         x                                              
         
  +         ,  I         J  g         h                                                                    9         :  W         X  u         v                                                         )         *  G         H  e          f   !         p"  #         #  $         $  %         %  &         &  (         (  #)         $)  A*         B*  _+         `+  },         ~,  -         -  .         .  /         /  0         0  2         2  13         23  O4         P4  m5         n5  6         6  7         7  8         8  9         9  ;         ;  !<         "<  ?=         @=  ]>         ^>  {?         |?  @         @  A         A  B         B  C         C  E         E  /F         0F  MG         :H  WI         XI  uJ         vJ  K         K  L         L  M         M  N         N  P         P  )Q         *Q  GR         HR  eS         fS  T         T  U         U  V         V  W         W  X         X  Z         Z  7[         8[  U\         V\  s]         t]  ^         ^  _         _  `         `  a         a  	c         
c  'd         (d  Ee         Fe  cf            ހ                      >  V          z              ΅                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               *!         @!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   q      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -K -X     K -հ-     հ-ܰ-    ܰ-/-p    /-G;-T    G;-c-4    c--    --8    -Eb-    Eb-x-	    x--8
    -3׳-
                   =         b      J      7   U   7   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               2!         >!  51                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 64 })  outputElements  5184    input0Elements  31104   input1Elements  24576   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     4                           #         b      J      7   M     operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       @                        ?  T          ~                                                .            G  -	         `	  
         
                                                                                                  b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     4                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  1728    input0Elements  1728    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       x         R  A          t  c                                  &  '         '  (         (  	*         *  5+         7+  a,         c,  -         -  .         .  /         /  1         1  =2         ?2  i3         k3  4         4  5         5  6         6  8         8  E9         G9  q:         s:  ;         ;  <         <  =         =  "?         #?  N@         O@  zA         {A  B         B  C         C  D         D  *F         +F  VG         WG  H         H  I         I  J         J  L         L  2M         3M  ^N         _N  O         O  P         P  Q         Q  S         S  :T         ;T  fU         gU  V         V  W         W  X         X  Z         Z  B[         C[  n\         o\  ]         ]  ^         ^  _         _  a         a  Jb         Kb  vc         wc  d         d  e         e  f         f  &h         'h  Ri         Si  ~j         j  k         k  l          n  +o         ,o  Wp         Xp  q         q  r         r  s         s  u         u  3v         4v  _w         `w  x         x  y         y  z         z  |         |  ;}         <}  g~         h~                                              C         D  o         p             Ǉ         ȇ                         K         L  w         x             ώ         Ў             '                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               &!         4!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     --    -<-x    <-J-|    J-~-\    ~--    --h    -ib-    ib-?-	    ?--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               A!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             $         %  I         J  n         o                                               "  >          a  }                                              =  B          l  q                                   :          ;   U!         V!  p"         q"  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       8         :  U         W  r         t                                              
         
            "  =         ?  Z         \  w         y                                                       
  %         '  B         D  _         a  |         ~                                                         *         ,  G          I   d!         d#  $         $  %         %  &         &  '         '  )         )  **         ,*  N+         P+  r,         t,  -         -  .         .  /         /  1         1  $2         &2  F3         H3  h4         j4  5         5  6         6  7         7  8         8  :         :  4;         6;  V<         X<  v=         x=  >         >  ?         ?  @         @  A         A  C         C  6D         8D  VE         XE  vF         xF  G          a  'b          Kb  Sc          wc  d          d  e          e  f         m  n         n  :o         ;o  Vp         Wp  rq         sq  r         r  s         s  t         t  u         u  v         v  x         x  6y         7y  Rz         Sz  n{         o{  |         |  }         }  ~         ~                                   2         3  N         O  j         k                                   ډ         ۉ                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               3!         J!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   r      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-[-     [--    -:-x    :-I-|    I-|-\    |-5-    5--P    -Wb-    Wb-c-	    c-a-D
    a--׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               +!         <!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                !  ;	         <	  V
         W
  q         I  c         d  ~                                                                   &  (          Q  S          |  ~                                    
           4   9!          c!  h"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p                     "  =         D  _         f                                              		         	  +
         2
  M         T  o         v                                                          ;         B  ]         d                                                         )         0  K         R  m         t                                     !         !  #         #  2$         7$  Q%         V%  p&         u&  '         '  (         (  )         )  +         +  6,         =,  ^-         e-  .         .  /         /  0         0  1         2  (3         /3  R4         Y4  |5         5  6         6  7         7  8         9  $:         +:  N;         U;  x<         <  =         =  >         >  ?         @  &A         -A  RB         YB  ~C         C  D         D  E         E  G         	G  .H         5H  OI           K  L          @L  YM          M  N          N  O           P  Q          n  o         o  9p         >p  Xq         ]q  wr         |r  s         s  t         t  u         u  v         v  x         x  1y         6y  Pz         Uz  o{         t{  |         |  }         }  ~         ~             
           )         .  H         M  g         l                        ć                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               -!         @!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 y
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-<-x    <-J-|    J-~-\    ~--    --h    -ib-    ib-?-	    ?--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               C!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             %         &  K         L  q         r                                               '  B          f                                                B  H          q  w                     
  $         %  ?          @   Z!         [!  u"         v"  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       9         :  V         W  s         t                                              
         
  !         "  >         ?  [         \  x         y                                              	         
  &         '  C         D  `         a  }         ~                                                         +         ,  H          I   e!         d#  ~$         $  %         %  &         &  '         '  )         )  +*         ,*  O+         P+  s,         t,  -         -  .         .  /         /  1         1  %2         &2  G3         H3  i4         j4  5         5  6         6  7         7  8         8  :         :  5;         6;  W<         X<  w=         x=  >         >  ?         ?  @         @  A         A  C         C  7D         8D  WE         XE  wF         xF  G          f  h          'h  /i          Si  [j          j  k          k  l         m  n         n  9o         ;o  Up         Wp  qq         sq  r         r  s         s  t         t  u         u  v         v  x         x  5y         7y  Qz         Sz  m{         o{  |         |  }         }  ~         ~                                   1         3  M         O  i         k                                   ى         ۉ                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               3!         J!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   q      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-[-     [--    -:-x    :-I-|    I-|-\    |-5-    5--P    -Wb-    Wb-c-	    c-a-D
    a--׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               *!         :!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                !  ;	         <	  V
         W
  q         I  c         d  ~                                                                   &  '          Q  R          |  }                                               4   :!          c!  i"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p                     "  <         D  ^         f                                              	         	  *
         2
  L         T  n         v                                                          :         B  \         d  ~                                                       (         0  J         R  l         t                                     !         !  #         #  3$         7$  R%         V%  q&         u&  '         '  (         (  )         )  +         +  5,         =,  ]-         e-  .         .  /         /  0         0  1         2  '3         /3  Q4         Y4  {5         5  6         6  7         7  8         9  #:         +:  M;         U;  w<         <  =         =  >         >  ?         @  %A         -A  QB         YB  }C         C  D         D  E         E  G         	G  -H         5H  PI          @Q  YR          R  S          S  T           U  V          @V  YW          n  o         o  :p         >p  Yq         ]q  xr         |r  s         s  t         t  u         u  v         v  x         x  2y         6y  Qz         Uz  p{         t{  |         |  }         }  ~         ~                        *         .  I         M  h         l                        Ň                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               -!         @!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 y
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    <     -,-p     ,--     --    -/5-    /5-C-    C-w-t    w-3-     3-;-X    ;--    -ub-p	    ub--x
    --
    -[-    [-ճ-\    ճ--<                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ?!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        /  4          ^  c                                                          A  =	          l	  h
          
                                              -         .  H            :         ;  U         V  p         q                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h         D%  I&         JL  uM         vM  N         N  O         O  P         P  %R         &R  QS         RS  }T         ~T  U         U  V         V  X         X  -Y         .Y  YZ         ZZ  [         [  \         \  ]         ]  	_         
_  5`         6`  aa         ba  b         b  c         c  d         d  f         f  =g         >g  ih         jh  i         i  j         j  k         k  m         m  En         Fn  qo         ro  p         p  q         q  r         r  !t         "t  Mu         Nu  yv         zv  w         w  x         x  y         y  ){         *{  U|         V|  }         }  ~         ~                        1         2  ]         ^                                              9         :  e         f                                              A         B  m         n                        Ɠ  ˔                        #         J  t         v             ̚         Κ             $         &  P         R  |         ~             ԡ         ֡              ,         .  X         Z                        ܨ         ި           
  4         6  `         b                                              <         >  h         j                                b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               @!         P!  1         1  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    b
         c
             /                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    b
         c
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            4     -,-`     ,-g-     g--    -;-    ;-J-    J-}-d    }-	-    	--p    -ub-    ub--
    --
    -ϰ-t    ϰ-'׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               2!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        5         6  P         Q  k         l                                       %          O  T          ~                                    +  A          j                                                 /!         0!  N"         O"  m#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                 @  I          p  y                     #  $         $  %         %  &         '  #(         *(  G)         N)  k*         r*  +         +  ,         ,  -         -  .         /  0         &0  C1         J1  g2         n2  3         3  4         4  5         5  6         6  8         "8  ?9         F9  c:         j:  ;         ;  <         <  =         =  >         >  @         @  ;A         BA  _B         fB  C         C  D         D  E         E  F         F  H         H  6I         ;I  WJ         \J  xK         }K  L         L  M         M  N         N  P         
P  ,Q         .Q  PR         RR  tS         vS  T         T  U         U  V         V  X         X  &Y         (Y  HZ         JZ  j[         l[  \         \  ]         ]  ^         ^  _         _  a         a  6b         8b  Xc         Zc  xd         zd  e         e  f         f  g         g  h         h  j         j  8k         :k  Xl         Zl  xm         zm  n         n  o         o  p         p  q         q  s         s  :t         ?t  [u         `u  |v         v  w         w  x         x  y         y   {         {  !|         &|  B}         G}  c~         h~                        Ɓ         ˁ                        )         .  J         O  k         p                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               D!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    b
         c
             ,      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     ;
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    <     -,-p     ,--     -#-    #-/5-    /5-C-    C-w-t    w-3-     3-;-X    ;--    -ub-p	    ub--x
    --
    -[-    [-ճ-\    ճ--<                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               A!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        /  5          ^  d                                                          A  <	          l	  g
          
                                              -         .  H            :         ;  U         V  p         q                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        D%  o&         JL  tM         vM  N         N  O         O  P         P  $R         &R  PS         RS  |T         ~T  U         U  V         V   X         X  ,Y         .Y  XZ         ZZ  [         [  \         \  ]         ]  _         
_  4`         6`  `a         ba  b         b  c         c  d         d  f         f  <g         >g  hh         jh  i         i  j         j  k         k  m         m  Dn         Fn  po         ro  p         p  q         q  r         r   t         "t  Lu         Nu  xv         zv  w         w  x         x  y         y  ({         *{  T|         V|  }         }  ~         ~                        0         2  \         ^                                              8         :  d         f                                              @         B  l         n             œ         Ɠ                        I          J  O          v  {                      Κ  ӛ                     &  Q         R  }         ~             ա         ֡             -         .  Y         Z                        ݨ         ި  	         
  5         6  a         b                                              =         >  i         j                                b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               B!         P!  1         1  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    b
         c
             0                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    b
         c
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            4     -,-`     ,-m-     m--    -;-    ;-J-    J-}-d    }-	-    	--p    -ub-    ub--
    --
    -հ-t    հ-'׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               6!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        5         6  P         Q  k         l                                       &          O  U          ~                                    +  @          j                                                 2!         3!  R"         S"  r#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                        	          0  9          `  i                     #  $         $  %         %  &         '  "(         *(  F)         N)  j*         r*  +         +  ,         ,  -         -  .         /  0         &0  B1         J1  f2         n2  3         3  4         4  5         5  6         6  8         "8  >9         F9  b:         j:  ;         ;  <         <  =         =  >         >  @         @  :A         BA  ^B         fB  C         C  D         D  E         E  F         F  H         H  7I         ;I  XJ         \J  yK         }K  L         L  M         M  N         N  	P         
P  -Q         .Q  QR         RR  uS         vS  T         T  U         U  V         V  X         X  'Y         (Y  IZ         JZ  k[         l[  \         \  ]         ]  ^         ^  _         _  a         a  7b         8b  Yc         Zc  yd         zd  e         e  f         f  g         g  h         h  j         j  9k         :k  Yl         Zl  ym         zm  n         n  o         o  p         p  q         q  s         s  ;t         ?t  \u         `u  }v         v  w         w  x         x  y         y  {         {  "|         &|  C}         G}  d~         h~                        ǁ         ˁ             	           *         .  K         O  l         p                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               B!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    b
         c
             -      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     <
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-y-     y--    -<-x    <-J-|    J-}-\    }-96-    96--h    -b-    b-	n-	    	n--h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               <!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             $         %  I         J  n         o                                               "  >          a  }                                              =  B          l  q                                   :          ;   U!         V!  p"         q"  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            (         *  R         T  |         ~                                   $         &  N	         P	  x
         z
                                             "  J         L  t         v                                              F         H  p         r                                              B         D  l         n               !         !  "          `1  u2          2  3          3  4          5  )6          P6  e7         K  L         L  N         N  1O         2O  [P         \P  Q         Q  R         R  S         S  U         U  -V         .V  WW         XW  X         X  Y         Y  Z         Z  [          \  )]         *]  S^         T^  }_         ~_  `         `  a         a  b         b  %d         &d  Oe         Pe  yf         zf  g         g  h         h  i         i  !k         "k  Kl         Ll  um         vm  n         n  o           6         8  `         b                        ޚ                    
  2         4  \         ^                        ڡ         ܡ             .         0  X         Z                        ֨         ب              *         ,  T         V  ~                      ү         ԯ             &         (  P         R  z         |             ζ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               U!         V!   2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     @
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                 w        Yܯ--0     --     -c-     c-B-P    B-c-    c--    -'׳-p                             b      *      7   I      operation.0     sync             :            2            b         reason  ExchangeBarrier value   1       invocations     1                                     b      *      7   I      operation.0     sync             :            2            b         reason  ExchangeBarrier value   3       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   5       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   2       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     --    -<-x    <-J-|    J-}-\    }-96-    96--h    -b-    b-	n-	    	n--h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               >!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             %         &  K         L  q         r                                               '  B          f                                                B  H          q  w                     
  $         %  ?          @   Z!         [!  u"         v"  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            )         *  S         T  }         ~                                   %         &  O	         P	  y
         z
                                   !         "  K         L  u         v                                              G         H  q         r                                              C         D  m         n               !         !  "          7  8          8  9          :  ;          @;  U<          |<  =         K  L         L  N         N  0O         2O  ZP         \P  Q         Q  R         R  S         S  U         U  ,V         .V  VW         XW  X         X  Y         Y  Z         Z  [          \  (]         *]  R^         T^  |_         ~_  `         `  a         a  b         b  $d         &d  Ne         Pe  xf         zf  g         g  h         h  i         i   k         "k  Jl         Ll  tm         vm  n         n  o           7         8  a         b                        ߚ           	         
  3         4  ]         ^                        ۡ         ܡ             /         0  Y         Z                        ר         ب             +         ,  U         V                        ӯ         ԯ             '         (  Q         R  {         |             ϶                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               Y!         Z!  2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     ?
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                 w        Yܯ--0     --     -c-     c-B-P    B-c-    c--    -'׳-p                             b      *      7   I      operation.0     sync             :            2            b         reason  ExchangeBarrier value   1       invocations     1                                     b      *      7   I      operation.0     sync             :            2            b         reason  ExchangeBarrier value   3       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   5       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   2       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,-7-     7--    -A5-    A5-C-    C-)w-l    )w-30-    30--x    -ib-	    ib--
    --
    -[-|    [-ճ-    ճ-׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     f            	!          F!  1          1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        -  .          Z  [                                                          ;  ?	          h	  l
          
                                              5         6  R         4  P         Q  m         n                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         8  E          n  {                                             T$  v%         x%  &         vL  M         M  N         N  O         O  Q         Q  )R         *R  MS         NS  qT         rT  U         U  V         V  W         W  Y         Y  %Z         &Z  I[         J[  m\         n\  ]         ]  ^         ^  _         _  `         `  !b         "b  Ec         Fc  id         jd  e         e  f         f  g         g  h         h  j         j  Ak         Bk  el         fl  m         m  n         n  o         o  p         p  r         r  =s         >s  at         bt  u         u  v         v  w         w  x         x  z         z  9{         :{  ]|         ^|  }         }  ~         ~                                   5         6  Y         Z  }         ~             Ň         Ƈ                        1         2  U         V  y         z                                            
  ,         .  P         R  t         v                                              (         *  L         N  p         r                        ܠ         ޠ              $         &  H         J  l         n                        ب         ڨ                       "  D         F  h         j                        ԰                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               G!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    d
         e
             7      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    d
         e
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,-7-     7--    -;-    ;-J-    J-}-l    }-a-    a--P    --    -b-h	    b--p
    --
    -ð-    ð--L                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     m            !          B!  1          1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        9         :  V         W  s         t                                                I  J          v  w                                    7          Z  t                                             #!         $!  H"         I"  m#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        $  %         %  &         &  (         (  ')         ()  M*         N*  s+         t+  ,         ,  -         -  .         .  0         0  11         21  W2         X2  }3         ~3  4         4  5         5  6         6  8         8  ;9         <9  a:         b:  ;         ;  <         <  =         =  >         >  @          @  EA         FA  kB         lB  C         C  D         D  E         E  G         G  (H         *H  NI         PI  tJ         vJ  K         K  L         L  M         M  O         O  2P         4P  XQ         ZQ  ~R         R  S         S  T         T  U         U  W         W  <X         >X  bY         dY  Z         Z  [         [  \         \  ]         ]   _         "_  F`         H`  la         na  b         b  c         c  d         d  f         
f  0g         2g  Xh         Zh  i         i  j         j  k         k  l         l   n         "n  Ho         Jo  np         pp  q         q  r         r  s         s  u          u  v          .v  +w          Tw  Qx          zx  wy          y  z         z  {         {  }         }  6~         8~  \         ^                        ΂         Ђ                        @         B  f         h                        ؊         ڊ                                b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               Y!         Z!  2         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    d
         e
             6                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     7
                   
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            8     -,-h     ,-7-     7--    -A5-    A5-C-    C-)w-l    )w-90-    90--x    -cb-	    cb--
    --
    -U-|    U-ճ-    ճ-׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     e            !          F!  1          1  ~=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        -  /          Z  \                                                          ;  >	          h	  k
          
                                              5         6  R         4  P         Q  m         n                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         F  S          |                !          !  "          #  +$         T$  w%         x%  &         vL  M         M  N         N  O         O  Q         Q  (R         *R  LS         NS  pT         rT  U         U  V         V  W         W   Y         Y  $Z         &Z  H[         J[  l\         n\  ]         ]  ^         ^  _         _  `         `   b         "b  Dc         Fc  hd         jd  e         e  f         f  g         g  h         h  j         j  @k         Bk  dl         fl  m         m  n         n  o         o  p         p  r         r  <s         >s  `t         bt  u         u  v         v  w         w  x         x  z         z  8{         :{  \|         ^|  }         }  ~         ~                                   4         6  X         Z  |         ~             ć         Ƈ                        0         2  T         V  x         z                                   	         
  -         .  Q         R  u         v                                              )         *  M         N  q         r                        ݠ         ޠ             %         &  I         J  m         n                        ٨         ڨ             !         "  E         F  i         j                        հ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               C!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    d
         e
             6      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    d
         e
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,-7-     7--    -;-    ;-J-    J-}-l    }-a-    a--P    --    -b-h	    b--p
    --
    -ð-    ð--L                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     l            !          B!  1          1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        9         :  V         W  s         t                                                I  K          v  x                                    6          Z  s                                                &!         '!  L"         M"  r#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        $  %         %  &         &   (         (  &)         ()  L*         N*  r+         t+  ,         ,  -         -  .         .  
0         0  01         21  V2         X2  |3         ~3  4         4  5         5  6         6  8         8  :9         <9  `:         b:  ;         ;  <         <  =         =  >         >  @          @  DA         FA  jB         lB  C         C  D         D  E         E  G         G  )H         *H  OI         PI  uJ         vJ  K         K  L         L  M         M  O         O  3P         4P  YQ         ZQ  R         R  S         S  T         T  U         U  W         W  =X         >X  cY         dY  Z         Z  [         [  \         \  ]         ]  !_         "_  G`         H`  ma         na  b         b  c         c  d         d  	f         
f  1g         2g  Yh         Zh  i         i  j         j  k         k  l         l  !n         "n  Io         Jo  op         pp  q         q  r         r  s         s  u         u  -v         .v  Sw         Tw  yx         zx  y         y  z          z  {          {  |          }  ~          8~  5          ^  [                      ς         Ђ                        A         B  g         h                        ي         ڊ                                b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               Y!         Z!  2         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    d
         e
             7                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     8
                   
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,-C-     C--    -;-x    ;-_J-|    _J-}-\    }-c3-    c3--h    -b-    b-m-	    m--h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     h            !          L!  1          1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        9         :  V         W  s         t            r                                    )  9          f  v                                              *  +          W  X                                 !         !  ("         )"  E#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                      '  G         N  n         u                                   
           1	         8	  X
         _
                                                       "  B         I  i         p                                              ,         3  S         Z  z                                                       =          D   d!         k!  "          #  $          $  %          &  '          @'  N(          z(  )         K  L         L  N         N  5O         >O  cP         lP  Q         Q  R         R  S         S  U         $U  IV         RV  wW         W  X         X  Y         Y  Z         [  '\         0\  S]         \]  ^         ^  _         _  `         `  b         b  /c         8c  Yd         bd  e         e  f         f  g         g  i         
i  +j         4j  Uk         ^k  l         l  m         m  n         ȑ                        0         4  T         X  x         |                      ę                        ,         0  P         T  t         x                                              (         ,  L         P  p         t                        ܪ                       $         (  H         L  l         p                        ز         ܲ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               N!         R!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     4
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            8     -,-h     ,-g-     g--    -5-    5-C-    C-v-l    v---    ---x    -9b-	    9b--
    --
    -m-|    m-ճ-    ճ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               5!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        =  W          z                                    1  $          H  ;          _  R	          v	  i
          
                                              7         8  \                      ;         <  X         Y  u         v                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                     	            .  *          S  O          x  t         V"  {#         }#  $         H  I         I  J         J  K         K  M         M  EN         GN  mO         oO  P         P  Q         Q  R         R  T         T  5U         7U  ]V         _V  W         W  X         X  Y         Y  Z         Z  %\         '\  M]         O]  u^         w^  _         _  `         `  a         a  c         c  =d         ?d  ee         ge  f         f  g         g  h         h  j         j  -k         /k  Ul         Wl  }m         m  n         n  o         o  p         p  r         r  Es         Gs  mt         ot  u         u  v         v  w         w  y         y  5z         7z  ]{         _{  |         |  }         }  ~         ~             %         '  M         O  u         w             Ņ         ǅ                        =         ?  e         g                        ݍ         ߍ             ,         .  S         U  {         }             ˔         ͔                        C         E  k         m                                              3         5  [         ]                        ӣ         գ             #         %  K         M  s         u             ê         Ū                        ;                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     m            !          H!  1          1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    d
         e
             6      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    f
         g
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            0     -,-X     ,-I-     I--    -;-x    ;-_J-|    _J-}-\    }-c3-    c3--h    -{b-    {b-m-	    m--h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     i            !          L!  1          1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        9         :  V         W  s         t            r                                    )  8          f  u                                              *  ,          W  Y                                 !         !  ("         )"  E#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            !         '  H         N  o         u                                              2	         8	  Y
         _
                                                       "  C         I  j         p                                              -         3  T         Z  {                                                       >          D   e!         k!  "          )  *          *  +          (,  6-          b-  p.          .  /         K  L         L  N         N  4O         >O  bP         lP  Q         Q  R         R  S         S  U         $U  HV         RV  vW         W  X         X  Y         Y  Z         [  &\         0\  R]         \]  ~^         ^  _         _  `         `  b         b  .c         8c  Xd         bd  e         e  f         f  g         g   i         
i  *j         4j  Tk         ^k  ~l         l  m         m  n         ȑ                        1         4  U         X  y         |                      ę             	           -         0  Q         T  u         x                                              )         ,  M         P  q         t                        ݪ                      %         (  I         L  m         p                        ٲ         ܲ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               L!         R!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     3
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            8     -,-h     ,-g-     g--    -5-    5-C-    C-w-l    w---    ---x    -3b-	    3b--
    --
    -m-|    m-ճ-    ճ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               3!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        =  V          z                                    1  #          H  :          _  Q	          v	  h
          
                                              ;         <  a           (         )  E         F  b         c                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                                !          1!  -"         V"  |#         }#  $         H  I         I  J         J  K         K  M         M  FN         GN  nO         oO  P         P  Q         Q  R         R  T         T  6U         7U  ^V         _V  W         W  X         X  Y         Y  Z         Z  &\         '\  N]         O]  v^         w^  _         _  `         `  a         a  c         c  >d         ?d  fe         ge  f         f  g         g  h         h  j         j  .k         /k  Vl         Wl  ~m         m  n         n  o         o  p         p  r         r  Fs         Gs  nt         ot  u         u  v         v  w         w  y         y  6z         7z  ^{         _{  |         |  }         }  ~         ~             &         '  N         O  v         w             ƅ         ǅ                        >         ?  f         g                        ލ         ߍ             -         .  T         U  |         }             ̔         ͔                        D         E  l         m                                              4         5  \         ]                        ԣ         գ             $         %  L         M  t         u             Ī         Ū                        <                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     l            !          D!  1          1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    d
         e
             7      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    g
         h
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            0     -,-X     ,-s-     s--    -G5-x    G5-C-|    C-/w-\    /w--    --h    -Eb-    Eb-m-	    m--h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               7!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        -  .          Z  [                                                          ;  ?	          h	  l
          
                                              5         6  R         4  P         Q  m         n                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                     %  C         J  h         o                                              !	         (	  F
         M
  k         r                                              $         +  I         P  n         u                                            	  '         .  L         S  q         x                                                *!         1!  O"         I  J         J  K         K  L         M  !N         (N  FO         MO  kP         rP  Q         Q  R         R  S         S  T         U  $V         +V  IW         PW  nX         uX  Y         Y  Z         Z  [         [  ]         	]  '^         .^  L_         S_  q`         x`  a         a  b         b  c         c  	e         e  /f         6f  Ug         \g  {h         h  i         i  j         j  k         k  m         m  ;n         Dn  co         lo  p         p  q         q  r         r  t         t  +u         4u  Sv         \v  {w         w  x         x  y         y  z         z  |         $|  C}         L}  k~         t~                      Ā                        3         <  [         d                        ӈ         ܈             #         ,  K         T  s         |                                    H  F          r  p                               >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     l            !          P!  1          1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     4
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-s-     s--    -;-x    ;-J-|    J-}-\    }-.-    .--h    -Kb-    Kb-n-	    n--h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               :!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        9         :  V         W  s         t                                                I  J          v  w                                    7          Z  t                                             #!         $!  H"         I"  m#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            (         *  R         T  |         ~                                   $         &  N	         P	  x
         z
                                             "  J         L  t         v                                              F         H  p         r                                              B         D  l         n               !          "  -#          T#  p$          $  %          %  &          '  9(         @G  iH         jH  I         I  J         J  K         K  "M         $M  LN         NN  vO         xO  P         P  Q         Q  R         R  T          T  HU         JU  rV         tV  W         W  X         X  Y         Y  [         [  D\         F\  n]         p]  ^         ^  _         _  `         `  b         b  @c         Bc  jd         ld  e         e  f         f  g         g  i         i  <j         >j  fk         hk  l         l  m           ͏         Ώ             !         "  K         L  u         v             ɖ         ʖ                        G         H  q         r             ŝ         Ɲ                        C         D  m         n                      ¤                        ?         @  i         j                                              ;                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     o            !          R!  1          1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     O
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-s-     s--    -G5-x    G5-C-|    C-/w-\    /w-"-    "--h    -?b-    ?b-m-	    m--h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               9!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        -  /          Z  \                                                          ;  >	          h	  k
          
                                              5         6  R         4  P         Q  m         n                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                     %  D         J  i         o                                              "	         (	  G
         M
  l         r                                               %         +  J         P  o         u                                            	  (         .  M         S  r         x                                                +!         1!  P"         I  J         J  K         K  L         M  "N         (N  GO         MO  lP         rP  Q         Q  R         R  S         S   U         U  %V         +V  JW         PW  oX         uX  Y         Y  Z         Z  [         [  ]         	]  (^         .^  M_         S_  r`         x`  a         a  b         b  c         c  
e         e  0f         6f  Vg         \g  |h         h  i         i  j         j  k         k  m         m  :n         Dn  bo         lo  p         p  q         q  r         r  t         t  *u         4u  Rv         \v  zw         w  x         x  y         y  z         z  |         $|  B}         L}  j~         t~                      Ā             
           2         <  Z         d                        ҈         ܈             "         ,  J         T  r         |            Ɲ  Ğ                                  D  B          n  l                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     k            !          L!  1          1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     3
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-s-     s--    -;-x    ;-J-|    J-}-\    }-.-    .--h    -Kb-    Kb-n-	    n--h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               :!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        9         :  V         W  s         t                                                I  K          v  x                                    6          Z  s                                                &!         '!  L"         M"  r#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            )         *  S         T  }         ~                                   %         &  O	         P	  y
         z
                                   !         "  K         L  u         v                                              G         H  q         r                                              C         D  m         n               !          `(  |)          )  *          *  ,          ),  E-          l-  .         @G  hH         jH  I         I  J         J  K         K  #M         $M  MN         NN  wO         xO  P         P  Q         Q  R         R  T          T  IU         JU  sV         tV  W         W  X         X  Y         Y  [         [  E\         F\  o]         p]  ^         ^  _         _  `         `  b         b  Ac         Bc  kd         ld  e         e  f         f  g         g  i         i  =j         >j  gk         hk  l         l  m           ̏         Ώ                       "  J         L  t         v             Ȗ         ʖ                        F         H  p         r             ĝ         Ɲ                        B         D  l         n                      ¤                        >         @  h         j                                              :                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     n            !          R!  1          1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     N
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-);-x    );-I-|    I-}-\    }-y-    y-}-h    }-b-    b-x-	    x--h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ?!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         1  M         N  j         k  	         	  
         
                                                         )          *  	          A             X  7          o  N            e                                    '           T   d!          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                        <         @  \         `  |                                                      	  
          
  <         @  \         `  |                                                                    <         @  \         `  |                                                                    <         @  \         `  |                        !         !  "         "  $         $  $%         +%  G&         N&  j'         q'  (         (  )         )  *         *  +         +  -          -  <.         C.  _/         f/  0         0  1         1  2         2  3         3  5         5  16         86  T7         [7  w8         ~8  9         9  :         :  ;         ;  =         
=  &>         ->  I?         P?  l@         s@  A         A  B         B  C         "D  >E         EE  aF         hF  G         G  H         I  J         K  L         &L  BM         IM  eN         lN  O         O  P         P  Q         Q  R         R  T         T  7U         >U  ZV         aV  }W         W  X         X  Y         Y  Z         Z  	\         \  ,]         3]  O^         V^  r_         y_  `         `  a         a  b         b  c         d  !e         (e  Df         Kf  gg          @  A          r  s                      ֆ  ׇ            	                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     `                       :!  1          1  g=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 s
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-y-     y--    -5-x    5-C-|    C-w-\    w--    --h    -ub-    ub-n-	    n--h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               <!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        =  W          z                                    1  $          H  ;          _  R	          v	  i
          
                                              7         8  \                      ;         <  X         Y  u         v                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            *         ,  V         X                                              2         4  ^	         `	  
         
                                   :         <  f         h                                              B         D  n         p                                               J         L  v         x                          !         !  &#         HI  rJ         tJ  K         K  L         L  M         M  "O         $O  NP         PP  zQ         |Q  R         R  S         S  T          U  *V         ,V  VW         XW  X         X  Y         Y  Z         Z  \         \  2]         4]  ^^         `^  _         _  `         `  a         a  c         c  :d         <d  fe         he  f         f  g         g  h         h  j         j  Bk         Dk  nl         pl  m         m  n         n  o         o  q          q  Jr         Lr  vs         xs  t         t  u         u  v         v  &x         (x  Ry         Ty  ~z         z  {         {  |         |  ~         ~  .         0  Z         \                        ރ           
           6         8  b         d                                              >         @  j         l                        ؚ              1          X  q                               >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               @!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     O
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     --    -/;-x    /;-I-|    I-}-\    }-s-    s--h    -'b-    'b-x-	    x--h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               =!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         1  M         N  j         k  	         	  
         
                                                         .          /            F  $          ]  ;          t  R            i                                    +           Y   h!          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                        =         @  ]         `  }                                                      	  
          
  =         @  ]         `  }                                                                    =         @  ]         `  }                                                                    =         @  ]         `  }                        !         !  "         "  $         $  %%         +%  H&         N&  k'         q'  (         (  )         )  *         *  +         +  -          -  =.         C.  `/         f/  0         0  1         1  2         2  3         3  5         5  26         86  U7         [7  x8         ~8  9         9  :         :  ;         ;  =         
=  '>         ->  J?         P?  m@         s@  A         A  B         B  C         "D  ?E         EE  bF         hF  G         G  H         I  J         K   L         &L  CM         IM  fN         lN  O         O  P         P  Q         Q  R         R  T         T  8U         >U  [V         aV  ~W         W  X         X  Y         Y  Z         Z  
\         \  -]         3]  P^         V^  s_         y_  `         `  a         a  b         b  c         d  "e         (e  Ef         Kf  hg          :  ;          l  m                      Ќ  э                               >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     a             !          >!  1          1  n=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 s
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-y-     y--    -#5-x    #5-C-|    C-w-\    w-_%-    _%--h    -ob-    ob-n-	    n--h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               <!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        =  V          z                                    1  #          H  :          _  Q	          v	  h
          
                                              ;         <  a           (         )  E         F  b         c                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            +         ,  W         X                                              3         4  _	         `	  
         
                                   ;         <  g         h                                              C         D  o         p                                               K         L  w         x                          !         !  '#         HI  sJ         tJ  K         K  L         L  M         M  #O         $O  OP         PP  {Q         |Q  R         R  S         S  T          U  +V         ,V  WW         XW  X         X  Y         Y  Z         Z  \         \  3]         4]  _^         `^  _         _  `         `  a         a  c         c  ;d         <d  ge         he  f         f  g         g  h         h  j         j  Ck         Dk  ol         pl  m         m  n         n  o         o  q          q  Kr         Lr  ws         xs  t         t  u         u  v         v  'x         (x  Sy         Ty  z         z  {         {  |         |  ~         ~  /         0  [         \                        ߃                      7         8  c         d                                              ?         @  k         l            ؟              1          X  q                      ؤ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               <!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     N
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-s-     s--    -;-x    ;-kJ-|    kJ-}-\    }--    --h    -3b-    3b-3-	    3--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               3!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        9         :  V         W  s         t            r                                    )  9          f  v                                              *  +          W  X                                 !         !  ("         )"  E#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ,         .  Z         \                                              @         B  n	         p	  
         
                        &         (  T         V                                              :         <  h         j                                             "  N         P  |         ~                        !         !  4"         6"  b#         d#  $         $  %         %  &         &  (         (  H)         J)  v*         x*  +         +  ,         ,   .         .  ./         0/  \0         ^0  1         1  2         2  3         3  5         5  B6         D6  p7         r7  8         8  9         9  :         :  (<         *<  V=         X=  >         >  ?         ?  @         @  B         B  <C         >C  jD         lD  E         E  F          G  H          H  I          I  K          :K  JL          xL  M         l  m         m  o          o  Mp         Np  {q         |q  r         r  s         s  u         u  3v         4v  aw         bw  x         x  y         y  z         z  |         |  G}         H}  u~         v~             р         Ҁ              -         .  [         \                                              A         B  o         p             ˍ         ̍             '                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               !         ,!  1         1  y=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
  b      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-;-x    ;-I-|    I-|-\    |-g-    g--h    -Kb-    Kb-x-	    x--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               D!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 7	         8	  T
         U
  q         S  o         p                                                                     >  8          k  e                                                            L   M!          y!  z"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            -         .  [         \                                              A         B  o	         p	  
         
                        '         (  U         V                                              ;         <  i         j                                   !         "  O         P  }         ~                        !         !  5"         6"  c#         d#  $         $  %         %  &         &  (         (  H)         J)  v*         x*  +         +  ,         ,   .         .  ./         0/  \0         ^0  1         1  2         2  3         3  5         5  B6         D6  p7         r7  8         8  9         9  :         :  (<         *<  V=         X=  >         >  ?         ?  @         @  B         B  <C         >C  jD         lD  E         E  F         F  G         G  "I         $I  PJ         RJ  ~K         K  L         L  M         M  O         
O  6P         8P  dQ         fQ  R         R  S         S  T         T  V         V  JW         LW  xX         zX  Y         Y  Z         Z  \         \  0]         2]  ^^         `^  _         _  `         `  a         a  c         c  Dd         Fd  re         te  f         f  g         g  h                                              :  *          ^  N         ʐ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               )!         :!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-m-     m--    -;-x    ;-kJ-|    kJ-}-\    }--    --h    -9b-    9b-3-	    3--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               1!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        9         :  V         W  s         t            r                                    )  8          f  u                                              *  ,          W  Y                                 !         !  ("         )"  E#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            -         .  [         \                                              A         B  o	         p	  
         
                        '         (  U         V                                              ;         <  i         j                                   !         "  O         P  }         ~                        !         !  5"         6"  c#         d#  $         $  %         %  &         &  (         (  I)         J)  w*         x*  +         +  ,         ,  .         .  //         0/  ]0         ^0  1         1  2         2  3         3  5         5  C6         D6  q7         r7  8         8  9         9  :         :  )<         *<  W=         X=  >         >  ?         ?  @         @  B         B  =C         >C  kD         lD  E         E  F          M  N          N  P          2P  BQ          pQ  R          R  S         l  m         m  o          o  Lp         Np  zq         |q  r         r  s         s  u         u  2v         4v  `w         bw  x         x  y         y  z         z  |         |  F}         H}  t~         v~             Ѐ         Ҁ              ,         .  Z         \                                              @         B  n         p             ʍ         ̍             &                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               !!         0!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
  a      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-;-x    ;-I-|    I-|-\    |-g-    g--h    -Qb-    Qb-x-	    x--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               @!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 7	         8	  T
         U
  q         S  o         p                                                           
          >  7          k  d                                                !           L   N!          y!  {"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ,         .  Z         \                                              @         B  n	         p	  
         
                        &         (  T         V                                              :         <  h         j                                             "  N         P  |         ~                        !         !  4"         6"  b#         d#  $         $  %         %  &         &  (         (  I)         J)  w*         x*  +         +  ,         ,  .         .  //         0/  ]0         ^0  1         1  2         2  3         3  5         5  C6         D6  q7         r7  8         8  9         9  :         :  )<         *<  W=         X=  >         >  ?         ?  @         @  B         B  =C         >C  kD         lD  E         E  F         F  G         G  #I         $I  QJ         RJ  K         K  L         L  M         M  	O         
O  7P         8P  eQ         fQ  R         R  S         S  T         T  V         V  KW         LW  yX         zX  Y         Y  Z         Z  \         \  1]         2]  _^         `^  _         _  `         `  a         a  c         c  Ed         Fd  se         te  f         f  g         g  h            r                      ʇ              މ                     ʐ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               +!         >!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-m-     m--    -5;-x    5;-I-|    I-}-\    }-k
-    k
--P    -]b-    ]b-c-	    c-m-D
    m-3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               0!         @!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         1  M         N  j         k  	         	  
         
                                                         )          *  	          A             X  7          o  N            e                                    '           T   d!          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            *         ,  V         X                                              2         4  ^	         `	  
         
                                   :         <  f         h                                              B         D  n         p                                               J         L  v         x                          !         !  &#         (#  R$         T$  ~%         %  &         &  '         '  )         )  .*         0*  Z+         \+  ,         ,  -         -  .         .  
0         0  61         81  b2         d2  3         3  4         4  5         5  7         7  >8         @8  j9         l9  :         :  ;         ;  <         <  >         >  F?         H?  r@         t@  A         A  B         B  C          R  S          S  T          T  U          V  W          LW  LX         h  (j         *j  Tk         Vk  l         l  m         m  n         n  p         p  0q         2q  \r         ^r  s         s  t         t  u         u  w         w  8x         :x  dy         fy  z         z  {         {  |         |  ~         ~  @         B  l         n                      Ă                        C         E  o         q             ǉ         ɉ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ,!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-y-     y--    -<-x    <-J-|    J-}-\    }-G-    G--h    -?b-    ?b-	-	    	--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               8!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        9         :  V         W  s         t                                                I  J          v  w                                    7          Z  t                                             #!         $!  H"         I"  m#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       :         <  X         Z  v         x                                              
         
  *         ,  H         J  f         h                                                                    8         :  V         X  t         v                                              
           (         *  F         H  d          f   !         #  $         $  &         &  :'         <'  Z(         \(  z)         |)  *         *  +         +  ,         ,  -         -  /         /  :0         <0  Z1         \1  z2         |2  3         3  4         4  5         5  6         6  8         8  :9         <9  Z:         \:  z;         |;  <         <  =         =  >         >  ?         ?  A         A  :B         <B  ZC         \C  zD         |D  E         F  G         G  H         H  J         J  ?K         @K  eL         fL  M         M  N         N  O         O  P         P  #R         $R  ES         GS  hT         jT  U         U  V         V  W         W  X         X  Z         Z  :[         <[  ]\         _\  ]         ]  ^         ^  _         _  `         `  b         b  %c         'c  Fd         Hd  ge         ie  f         f  g         g  h          i  j          k  l          Bl  Tm          m  n          n  o                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               "!         ,!  1         1  ~=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   3      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-g-     g--    -5;-x    5;-I-|    I-}-\    }-k
-    k
--P    -cb-    cb-c-	    c-m-D
    m-3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               /!         >!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         1  M         N  j         k  	         	  
         
                                                         .          /            F  $          ]  ;          t  R            i                                    +           Y   h!          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            +         ,  W         X                                              3         4  _	         `	  
         
                                   ;         <  g         h                                              C         D  o         p                                               K         L  w         x                          !         !  '#         (#  S$         T$  %         %  &         &  '         '  )         )  /*         0*  [+         \+  ,         ,  -         -  .         .  0         0  71         81  c2         d2  3         3  4         4  5         5  7         7  ?8         @8  k9         l9  :         :  ;         ;  <         <  >         >  G?         H?  s@         t@  A         A  B         B  C          zX  zY          Y  Z          Z  [          \  ]          2]  2^         h  )j         *j  Uk         Vk  l         l  m         m  n         n  p         p  1q         2q  ]r         ^r  s         s  t         t  u         u  w         w  9x         :x  ey         fy  z         z  {         {  |         |  ~         ~  A         B  m         n             Â         Ă                        D         E  p         q             ȉ         ɉ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               .!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-s-     s--    -<-x    <-J-|    J-}-\    }-}-    }--h    -Eb-    Eb-	-	    	--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               4!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        9         :  V         W  s         t                                                I  K          v  x                                    6          Z  s                                                &!         '!  L"         M"  r#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       ;         <  Y         Z  w         x                                              
         
  +         ,  I         J  g         h                                                                    9         :  W         X  u         v                                                         )         *  G         H  e          f   !         #  $         $  &         &  ;'         <'  [(         \(  {)         |)  *         *  +         +  ,         ,  -         -  /         /  ;0         <0  [1         \1  {2         |2  3         3  4         4  5         5  6         6  8         8  ;9         <9  [:         \:  {;         |;  <         <  =         =  >         >  ?         ?  A         A  ;B         <B  [C         \C  {D         |D  E         F  G         G  H         H  J         J  >K         @K  dL         fL  M         M  N         N  O         O  P         P  "R         $R  FS         GS  iT         jT  U         U  V         V  W         W  X         X  Z         Z  ;[         <[  ^\         _\  ]         ]  ^         ^  _         _  `         `  b         b  &c         'c  Gd         Hd  he         ie  f         f  g         g  h          o  q          :q  Lr          xr  s          s  t          t  v                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               $!         0!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   4      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,-m-     m--    -;-    ;-wJ-    wJ-}-d    }-A-    A--p    -{b-    {b--
    --
    --t    -3׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               1!         >!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        9         :  V         W  s         t            r                                    )  9          f  v                                              *  +          W  X                                 !         !  ("         )"  E#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         X  H          v  f                                             %  1&         2&  ['         \'  (         (  )         )  *         *  ,         ,  --         .-  W.         X.  /         /  0         0  1         1  2          3  )4         *4  S5         T5  }6         ~6  7         7  8         8  9         9  %;         &;  O<         P<  y=         z=  >         >  ?         ?  @         @  !B         "B  KC         LC  uD         vD  E         E  F         F  G         G  I         I  FJ         HJ  pK         rK  L         L  M         M  N         N  P         P  BQ         DQ  lR         nR  S         S  T         T  U         U  W         W  >X         @X  hY         jY  Z         Z  [         [  \         \  ^         ^  :_         <_  d`         f`  a         a  b         b  c         c  e         e  6f         8f  `g         bg  h         h  i         i  j         j  l         
l  2m         4m  \n         ^n  o         o  p         p  q         q  s         s  .t         0t  Xu         Zu  v         v  w         w  x         x   z         z  *{         ,{  T|         V|  ~}         }  ~         ~                        &         (  P         R  z         |             Ά         І             "         $  L         N  v         x             ʍ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               9!         R!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    d
         e
             7      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     H
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-s-     s--    -;-x    ;-I-|    I-|-\    |--    --P    -ib-    ib-c-	    c-[-D
    [-3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               5!         @!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 7	         8	  T
         U
  q         S  o         p                                                                     >  8          k  e                                                            L   M!          y!  z"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p                        ?         @  _         `                                                        	  
          
  ?         @  _         `                                                                      ?         @  _         `                                                                      ?         @  _         `                          !         !  "         "  #          $  %          %  >&         @&  ^'         `'  ~(         (  )         )  *         *  +         +  ,          -  .          .  >/         @/  ^0         `0  ~1         1  2         2  3         3  4         4  5          6  7          7  >8         @8  ^9         `9  ~:         :  ;         ;  <         <  =         =  >          ?  @          @  >A         G  
I         I  *J         +J  JK         KK  jL         kL  M         M  N         N  O         O  P         P  
R         R  *S         +S  JT         KT  jU         kU  V         V  W         W  X         X  Y         Y  
[         [  *\         +\  J]         K]  j^         k^  _         _  `         `  a         a  b         b  d         d  6e         7e  Xf         Yf  zg         {g  h           y  7z          cz  z{          {  |          |   ~          ,~  C                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               1!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 w
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,-y-     y--    -;-    ;-wJ-    wJ-}-d    }-G-    G--p    -{b-    {b--
    --
    --t    -3׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               5!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        9         :  V         W  s         t            r                                    )  8          f  u                                              *  ,          W  Y                                 !         !  ("         )"  E#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                 *            H  8           f   V!         %  0&         2&  Z'         \'  (         (  )         )  *         *  ,         ,  ,-         .-  V.         X.  /         /  0         0  1         1  2          3  (4         *4  R5         T5  |6         ~6  7         7  8         8  9         9  $;         &;  N<         P<  x=         z=  >         >  ?         ?  @         @   B         "B  JC         LC  tD         vD  E         E  F         F  G         G  I         I  GJ         HJ  qK         rK  L         L  M         M  N         N  P         P  CQ         DQ  mR         nR  S         S  T         T  U         U  W         W  ?X         @X  iY         jY  Z         Z  [         [  \         \  ^         ^  ;_         <_  e`         f`  a         a  b         b  c         c  e         e  7f         8f  ag         bg  h         h  i         i  j         j  	l         
l  3m         4m  ]n         ^n  o         o  p         p  q         q  s         s  /t         0t  Yu         Zu  v         v  w         w  x         x  z         z  +{         ,{  U|         V|  }         }  ~         ~                        '         (  Q         R  {         |             φ         І             #         $  M         N  w         x             ˍ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               9!         R!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    d
         e
             6      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     I
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-m-     m--    -;-x    ;-I-|    I-|-\    |--    --P    -ib-    ib-c-	    c-[-D
    [-3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               2!         >!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 7	         8	  T
         U
  q         S  o         p                                                           
          >  7          k  d                                                !           L   N!          y!  {"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p                        >         @  ^         `  ~                                                      	  
          
  >         @  ^         `  ~                                                                    >         @  ^         `  ~                                                                    >         @  ^         `  ~                        !         !  "         "  #          $  %          %  ?&         @&  _'         `'  (         (  )         )  *         *  +         +  ,          -  .          .  ?/         @/  _0         `0  1         1  2         2  3         3  4         4  5          6  7          7  ?8         @8  _9         `9  :         :  ;         ;  <         <  =         =  >          ?  @          @  ?A         G  	I         I  )J         +J  IK         KK  iL         kL  M         M  N         N  O         O  P         P  	R         R  )S         +S  IT         KT  iU         kU  V         V  W         W  X         X  Y         Y  	[         [  )\         +\  I]         K]  i^         k^  _         _  `         `  a         a  b         b  d         d  5e         7e  Wf         Yf  yg         {g  h          o              Ɂ                      8  O          {                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               1!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 w
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,-y-     y--    -5-    5-D-    D-kw-l    kw-12-    12--x    -ob-	    ob--
    -#-
    #-s-|    s-ճ-    ճ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               3!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        -  .          Z  [                                                          ;  <	          h	  i
          
                                              5         6  R         4  W         X  {         |                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       :  K          t                                   $  %         K  L         L  N         N  EO         FO  mP         nP  Q         Q  R         R  S         S  U         U  5V         6V  ]W         ^W  X         X  Y         Y  Z         Z  [         [  %]         &]  M^         N^  u_         v_  `         `  a         a  b         b  d         d  =e         >e  ef         ff  g         g  h         h  i         i  k         k  -l         .l  Um         Vm  }n         ~n  o         o  p         p  q         q  s         s  Et         Ft  mu         nu  v         v  w         w  x         x  z         z  5{         6{  ]|         ^|  }         }  ~         ~                        %         &  M         N  u         v             ņ         Ɔ                        =         >  e         f                        ݎ         ގ             -         .  T         V  |         ~             ̕         Ε                        D         F  l         n                                              4         6  \         ^                        Ԥ         ֤             $         &  L         N  t         v             ī         ƫ                        <         >  d         f                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               4!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    d
         e
             7      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    d
         e
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     h
            4     -,-`     ,-y-     y--    -a<-    a<-J-    J-I~-d    I~--    --p    -b-    b-!-
    !-/-
    /-ᰳ-t    ᰳ-9׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               6!         >!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        9         :  V         W  s         t                                                I  J          v  w                                    2          Z  o                                  7          8   g!         h!  "         "  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                     D  O          x                                   $  8%         :%  [&         \&  }'         ~'  (         (  )         )  *         *  ,         ,  '-         (-  I.         J.  k/         l/  0         0  1         1  2         2  3         3  5         5  76         86  Y7         Z7  {8         |8  9         9  :         :  ;         ;  =         =  %>         &>  G?         H?  i@         j@  A         A  B         B  C         C  D         D  F         F  5G         6G  VH         XH  xI         zI  J         J  K         K  L         L  "N         $N  UO         WO  P         P  Q         Q  R         R  !T         #T  TU         VU  V         V  W         W  X         X  'Z         (Z  ][         ^[  \         \  ]         ]  ^          _  5`         6`  ka         la  b         b  c         c  e         e  ?f         @f  sg         tg  h         h  i         i  k         k  Cl         Dl  wm         xm  n         n  o         o   q         q  "r         $r  Ds         Fs  ft         ht  u         u  v         v  w         w  x         x  z         z  2{         4{  T|         V|  v}         x}  ~         ~             ܀         ހ                        "  B         D  d         f                        ʈ         ̈                        0                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               >!         R!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    d
         e
             B      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     7
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     --    -}5-    }5-D-    D-ew-l    ew-12-    12--x    -ob-	    ob--
    -#-
    #-y-|    y-ճ-    ճ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               7!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        -  /          Z  \                                                          ;  =	          h	  j
          
                                              5         6  R         4  V         W  y         z                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         "  3          \  m            	          	  
          
           $  %         K  L         L  N         N  DO         FO  lP         nP  Q         Q  R         R  S         S  U         U  4V         6V  \W         ^W  X         X  Y         Y  Z         Z  [         [  $]         &]  L^         N^  t_         v_  `         `  a         a  b         b  d         d  <e         >e  df         ff  g         g  h         h  i         i  k         k  ,l         .l  Tm         Vm  |n         ~n  o         o  p         p  q         q  s         s  Dt         Ft  lu         nu  v         v  w         w  x         x  z         z  4{         6{  \|         ^|  }         }  ~         ~                        $         &  L         N  t         v             Ć         Ɔ                        <         >  d         f                        ܎         ގ             ,         .  U         V  }         ~             ͕         Ε                        E         F  m         n                                              5         6  ]         ^                        դ         ֤             %         &  M         N  u         v             ū         ƫ                        =         >  e         f                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               6!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    d
         e
             6      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    d
         e
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     g
            4     -,-`     ,--     --    -a<-    a<-J-    J-I~-d    I~--    --p    -b-    b-!-
    !-/-
    /-ᰳ-t    ᰳ-9׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               8!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        9         :  V         W  s         t                                                I  K          v  x                                    1          Z  n                                  4          5   e!         f!  "         "  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                     H  S           |   !          !  "          "  #         $  9%         :%  Z&         \&  |'         ~'  (         (  )         )  *         *  ,         ,  &-         (-  H.         J.  j/         l/  0         0  1         1  2         2  3         3  5         5  66         86  X7         Z7  z8         |8  9         9  :         :  ;         ;  =         =  $>         &>  F?         H?  h@         j@  A         A  B         B  C         C  D         D  F         F  4G         6G  WH         XH  yI         zI  J         J  K         K  L         L  #N         $N  VO         WO  P         P  Q         Q  R         R  "T         #T  UU         VU  V         V  W         W  X         X  &Z         (Z  \[         ^[  \         \  ]         ]  ^          _  4`         6`  ja         la  b         b  c         c  
e         e  >f         @f  rg         tg  h         h  i         i  k         k  Bl         Dl  vm         xm  n         n  o         o  q         q  #r         $r  Es         Fs  gt         ht  u         u  v         v  w         w  x         x  z         z  3{         4{  U|         V|  w}         x}  ~         ~             ݀         ހ              !         "  C         D  e         f                        ˈ         ̈                        1                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               <!         R!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    d
         e
             D      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     8
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-;-x    ;-J-|    J-}-\    }-4-    4--h    -b-    b-n-	    n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               C!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         ?         @  _         `                                                            8  =          u  z                                              9  :          f  g                                 !         !  7"         8"  T#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            $         &  J         L  p         r                                            
  .	         0	  T
         V
  z         |                                              8         :  ^         `                                                         B         D  h         j                                               &          (   L!         N!  r"          x2  3          3  4          4  6          A6  X7          7  8         K  L         L  M          N  %O         &O  KP         LP  qQ         rQ  R         R  S         S  T         T  	V         
V  /W         0W  UX         VX  {Y         |Y  Z         Z  [         [  \         \  ^         ^  9_         :_  _`         ``  a         a  b         b  c         c  d         d  f         f  Cg         Dg  ih         jh  i         i  j         j  k         k  m         m  'n         (n  Mo           ?         @  e         f                        י         ؙ             #         $  I         J  o         p                                              -         .  S         T  y         z             Ũ         ƨ                        7         8  ]         ^                        ϰ         а                        A         B  g                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               D!         J!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     )
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     -/-    /-55-    55-C-    C-w-l    w-5-    5--x    -b-	    b--
    -#-
    #-y-|    y-ճ-    ճ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               A!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        =  O          z                                    1            H  +          _  B	          v	  Y
          
  p                                                       4                      ;         <  X         Y  u         v                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                        $            H  @          l  d                     8"  Y#         [#  |$         ~$  %         %  &         &  '         '  )         )  0*         2*  T+         V+  x,         z,  -         -  .         .  /         /  1         
1  ,2         .2  P3         R3  t4         v4  5         5  6         6  7         7  9         9  (:         *:  L;         N;  p<         r<  =         =  >         >  ?         ?   A         A  $B         &B  HC         JC  lD         nD  E         E  F         F  G         -l  Nm         Pm  qn         sn  o         o  p         p  q         q  s         s  %t         't  Iu         Ku  mv         ov  w         w  x         x  y         y  z         z  !|         #|  E}         G}  i~         k~                        Ձ         ׁ                        A         C  e         g                        щ         Ӊ           ړ                       "  D         F  h         j                        ԛ         ֛                        @         B  d         f                        У         ң                        <         >  `         b                        ̫         Ϋ                        8         :  \         ^                        ȳ         ʳ                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               @!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    d
         e
             .      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    d
         e
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     W
            0     -,-X     ,--     -/-    /-;-x    ;-J-|    J-}-\    }-4-    4--h    -b-    b-n-	    n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               C!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                       !  A         B  b         c                                                            =  A          z  ~                                              >  @          k  m                                   !          !  <"         ="  Y#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            %         &  K         L  q         r                                   	         
  /	         0	  U
         V
  {         |                                              9         :  _         `                                                         C         D  i         j                                              '          (   M!         N!  s"          8  9          
:  !;          M;  d<          <  =          =  >         K  L         L  M          N  $O         &O  JP         LP  pQ         rQ  R         R  S         S  T         T  V         
V  .W         0W  TX         VX  zY         |Y  Z         Z  [         [  \         \  ^         ^  8_         :_  ^`         ``  a         a  b         b  c         c  d         d  f         f  Bg         Dg  hh         jh  i         i  j         j  k         k   m         m  &n         (n  Lo           >         @  d         f                        ֙         ؙ             "         $  H         J  n         p                                              ,         .  R         T  x         z             Ĩ         ƨ                        6         8  \         ^                        ΰ         а                        @         B  f                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               F!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     (
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     -5-    5-;5-    ;5-C-    C-#w-l    #w-5-    5--x    -b-	    b--
    -#-
    #-y-|    y-ճ-    ճ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               E!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        =  N          z                                    1            H  *          _  A	          v	  X
          
  o                                                       9           (         )  E         F  b         c                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                              	  
          D
  <         8"  Z#         [#  }$         ~$  %         %  &         &  '         '  )         )  1*         2*  U+         V+  y,         z,  -         -  .         .  /         /  	1         
1  -2         .2  Q3         R3  u4         v4  5         5  6         6  7         7  9         9  ):         *:  M;         N;  q<         r<  =         =  >         >  ?         ?  A         A  %B         &B  IC         JC  mD         nD  E         E  F         F  G         -l  Om         Pm  rn         sn  o         o  p         p  q         q  s         s  &t         't  Ju         Ku  nv         ov  w         w  x         x  y         y  z         z  "|         #|  F}         G}  j~         k~                        ց         ׁ                        B         C  f         g                        ҉         Ӊ           ړ             !         "  E         F  i         j                        ՛         ֛                        A         B  e         f                        ѣ         ң                        =         >  a         b                        ͫ         Ϋ                        9         :  ]         ^                        ɳ         ʳ                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               B!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    d
         e
             /      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    d
         e
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     X
            0     -,-X     ,--     -/-    /-_5-x    _5-C-|    C-Gw-\    Gw--    --h    -b-    b-n-	    n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               C!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        -  .          Z  [                                                
          ;  7	          h	  d
          
                                              5         6  R         4  P         Q  m         n                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            "         $  F         H  j         l                                              	          	  B
         D
  f         h                                                         >         @  b         d                                                         :         <  ^         `                                              !         !  6"          ;  <          <  =          $>  0?          \?  h@          @  A         .I  PJ         RJ  tK         vK  L         L  M         M  N         N  P         P  (Q         *Q  LR         NR  pS         rS  T         T  U         U  V         V   X         X  $Y         &Y  HZ         JZ  l[         n[  \         \  ]         ]  ^         ^  _         _  #a         $a  Ib         Jb  oc         pc  d         d  e         e  f         f  h         h  -i         .i  Sj         Tj  yk         m  n         n  p         p  <q         =q  `r         ar  s         s  t         t  u         u  v         v  x         x  8y         9y  \z         ]z  {         {  |         |  }         }  ~         ~             4         5  X         Y  |         }             ą         Ņ                        0         1  T         U  x         y                                                      >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               C!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     <
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-;-x    ;-}J-|    }J-}-\    }--    --h    -b-    b--n-	    -n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               <!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        9         :  V         W  s         t                                                I  J          v  w                                    /          Z  l                                             !         !  ("         )"  E#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            $         &  J         L  p         r                                            
  .	         0	  T
         V
  z         |                                              8         :  ^         `                                                         B         D  h         j                                               &          (   L!         N!  r"         t"  #         #  $         $  %         %  '         '  1(         2(  W)         X)  }*         ~*  +         +  ,         ,  -         -  /         /  ;0         <0  a1         b1  2          >  ?          ?  @          @   B          ,B  4C          `C  hD         _j  k         k  l         l  m         m  n         n  p         p  Aq         Cq  gr         ir  s         s  t         t  u         u  v         w  %x         'x  Ky         My  qz         sz  {         {  |         |  }         }  	           /         1  U         W  {         }             Ǆ         Ʉ                        9         ;  _         a                        ь         ӌ                      	  .         /  T         U  z         {             ƕ         Ǖ                        8         9  ^         _                        Н         ѝ                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               =!         J!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     O
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-_5-x    _5-C-|    C-Gw-\    Gw--    --h    -{b-    {b-n-	    n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               C!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        -  /          Z  \                                                	          ;  6	          h	  c
          
                                              5         6  R         4  P         Q  m         n                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            #         $  G         H  k         l                                              	          	  C
         D
  g         h                                                         ?         @  c         d                                                         ;         <  _         `                                              !         !  7"          A  B          C  D          <D  HE          tE  F          F  G         .I  QJ         RJ  uK         vK  L         L  M         M  N         N  P         P  )Q         *Q  MR         NR  qS         rS  T         T  U         U  V         V  X         X  %Y         &Y  IZ         JZ  m[         n[  \         \  ]         ]  ^         ^  _         _  "a         $a  Hb         Jb  nc         pc  d         d  e         e  f         f  h         h  ,i         .i  Rj         Tj  xk         m  n         n  p         p  ;q         =q  _r         ar  s         s  t         t  u         u  v         v  x         x  7y         9y  [z         ]z  {         {  |         |  }         }  ~         ~             3         5  W         Y  {         }             Å         Ņ                        /         1  S         U  w         y                                                      >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ?!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     ;
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-;-x    ;-J-|    J-}-\    }--    --h    -b-    b--n-	    -n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               >!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        9         :  V         W  s         t                                                I  K          v  x                                    .          Z  k                                             !         !  ,"         -"  J#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            %         &  K         L  q         r                                   	         
  /	         0	  U
         V
  {         |                                              9         :  _         `                                                         C         D  i         j                                              '          (   M!         N!  s"         t"  #         #  $         $  %         %  
'         '  0(         2(  V)         X)  |*         ~*  +         +  ,         ,  -         -  /         /  :0         <0  `1         b1  2          D  E          E  F          F  H          0H  8I          dI  lJ         _j  k         k  l         l  m         m  n         n  p         p  Bq         Cq  hr         ir  s         s  t         t  u         u   w         w  &x         'x  Ly         My  rz         sz  {         {  |         |  }         }  
           0         1  V         W  |         }             Ȅ         Ʉ                        :         ;  `         a                        Ҍ         ӌ                      	  -         /  S         U  y         {             ŕ         Ǖ                        7         9  ]         _                        ϝ         ѝ                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               A!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     N
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-;;-x    ;;-I-|    I-#}-\    #}-+-    +--h    -ob-    ob-x-	    x--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               E!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         1  M         N  j         k  	         	  
         
           q                                                                     7  &          N  =          e  T          |  k                                    %           J   b!          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                      "  B         D  d         f                                              	         	  0
         2
  R         T  t         v                                                          @         B  b         d                                                         .         0  P         R  r         t                                     !         "  1#         3#  S$          U$  b%          %  &          &  '          '  )          -)  :*         sG  H         H  I         I  J         J  K         K  M         M  =N         ?N  _O         aO  P         P  Q         Q  R         R  S         S  	U         U  +V         -V  MW         OW  oX         qX  Y         Y  Z         Z  [         [  \         \  ^         ^  ;_         =_  ]`         _`  a         a  b         b  c         c  d         d  f         	f  )g         +g  Kh         Mh  mi         oi  j         j  k         k  l         $  D         F  f         h                        ̕         Ε                        2         4  T         V  v         x                        ܞ         ޞ                        "  B         D  d         f                        ʦ         ̦                        0         2  R         T  t                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               6!         B!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 s
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-)5-x    )5-C-|    C-w-\    w-C-    C--h    -{b-    {b--n-	    -n--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               <!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        =  O          z                                    1            H  +          _  B	          v	  Y
          
  p                                                       4                      ;         <  X         Y  u         v                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            &         (  N         P  v         x                                              >	         @	  f
         h
                                              .         0  V         X  ~                                                        F         H  n         p                                              6         8  ^          `   !         !  "         "  $         $  :%         <%  b&         d&  '         '  (         (  )         )  +         +  *,         ,,  R-         T-  z.         |.  /         /  0         0  1         1  3         3  B4         D4  j5         l5  6         6  7         7  8         8  
:         :  2;         4;  Z<         \<  =         =  >         >  ?         ?  @         @  "B         $B  JC         LC  rD         tD  E         E  F          U`  ia          a  b          b  c          d  2e          ae  uf         Bm  hn         jn  o         o  p         p  q         q  s         
s  0t         2t  Xu         Zu  v         v  w         w  x         x  y         y   {         "{  H|         J|  p}         r}  ~         ~                                   8         :  `         b                        ؇         ڇ              (         *  P         R  x         z             Ȏ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               <!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     O
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-A;-x    A;-I-|    I-)}-\    )}-+-    +--h    -ub-    ub-x-	    x--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               A!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         1  M         N  j         k  	         	  
         
           v                                              )          *            A  /          X  F          o  ]            t                                    .           T   k!          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            !         "  C         D  e         f                                              	         	  1
         2
  S         T  u         v                                                          A         B  c         d                                                         /         0  Q         R  s         t                                     !         "  2#         3#  T$          c*  p+          +  ,          ,  -          .  /          ;/  H0         sG  H         H  I         I  J         J  K         K  M         M  >N         ?N  `O         aO  P         P  Q         Q  R         R  S         S  
U         U  ,V         -V  NW         OW  pX         qX  Y         Y  Z         Z  [         [  \         \  ^         ^  <_         =_  ^`         _`  a         a  b         b  c         c  d         d  f         	f  *g         +g  Lh         Mh  ni         oi  j         j  k         k  l         $  E         F  g         h                        ͕         Ε                        3         4  U         V  w         x                        ݞ         ޞ              !         "  C         D  e         f                        ˦         ̦                        1         2  S         T  u                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               8!         F!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 s
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-55-x    55-C-|    C-w-\    w-I-    I--h    -ub-    ub--n-	    -n--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               >!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        =  N          z                                    1            H  *          _  A	          v	  X
          
  o                                                       9           (         )  E         F  b         c                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            '         (  O         P  w         x                                              ?	         @	  g
         h
                                              /         0  W         X                                                          G         H  o         p                                              7         8  _          `   !         !  "         "  $         $  ;%         <%  c&         d&  '         '  (         (  )         )  +         +  +,         ,,  S-         T-  {.         |.  /         /  0         0  1         1  3         3  C4         D4  k5         l5  6         6  7         7  8         8  :         :  3;         4;  [<         \<  =         =  >         >  ?         ?  @         @  #B         $B  KC         LC  sD         tD  E         E  F          f  g          g  h          *i  >j          mj  k          k  l         Bm  in         jn  o         o  p         p  q         q  	s         
s  1t         2t  Yu         Zu  v         v  w         w  x         x  y         y  !{         "{  I|         J|  q}         r}  ~         ~                                   9         :  a         b                        ه         ڇ             )         *  Q         R  y         z             Ɏ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               :!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     N
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     --    -Y5-x    Y5-C-|    C-Aw-\    Aw--    --h    -Qb-    Qb-x-	    x--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               =!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        -  .          Z  [                                                
          ;  7	          h	  d
          
                                              5         6  R         4  P         Q  m         n                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                 @  >          p  n                     "  $         $  *%         ,%  J&         L&  j'         l'  (         (  )         )  *         *  +         +  
-         -  *.         ,.  J/         L/  j0         l0  1         1  2         2  3         3  4         4  
6         6  *7         ,7  J8         L8  j9         l9  :         :  ;         ;  <         <  =         =  
?         ?  *@         ,@  JA         LA  jB         lB  C         C  D         D  E         AF  _G         aG  H         H  I         I  J         J  K         K  L         M  N         !N  ?O         AO  _P         aP  Q         Q  R         R  S         S  T         T  U         V  W         !W  ?X         AX  _Y         aY  Z         Z  [         [  \         \  ]         ]  ^         _  `         !`  ?a         Aa  _b         ab  c         c  d         d  e         e  f         f  g         h  i         i  k         k  1l         2l  Qm         Rm  qn         rn  o         o  p         p  q         q  r         r  t         t  1u         2u  Qv         Rv  qw         rw  x         x  y         y  z         z  {         {  }         }  1~         2~  Q         R  q         r                        у         ҃                        1         2  Q         R  q                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               -!         4!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 s
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-;-x    ;-I-|    I-|-\    |-i-    i--h    -cb-    cb-x-	    x--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               >!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 7	         8	  T
         U
  q         S  o         p                                                                     >  @          k  m                                                            L   M!          y!  z"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            )         *  S         T  }         ~                                   %         &  O	         P	  y
         z
                                   !         "  K         L  u         v                                              G         H  q         r                                              C         D  m         n               !         !  "          "  #         $  <%         >%  f&         h&  '         '  (         (  )         )  +         +  8,         :,  b-         d-  .         .  /         /  0         0  
2         2  43         63  ^4         `4  5         5  6         6  7         7  9         9  0:         2:  Z;         \;  <         <  =         =  >         >  @         @  ,A         .A  VB         XB  C         C  D         D  E         E  F          G  (H         :H  bI         dI  J         J  K         K  L         L  
N         N  4O         6O  ^P         `P  Q         Q  R         R  S         S  U         U  0V         2V  ZW         \W  X         X  Y         Y  Z         Z  \         \  ,]         .]  V^         X^  _         _  `         `  a         a  b          c  (d         *d  Re         Te  |f         ~f  g         g  h          h  i          i  j          k  l          2l  &m         ʎ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               /!         B!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-y-     y--    -Y5-x    Y5-C-|    C-Aw-\    Aw--    --h    -Wb-    Wb-x-	    x--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               9!         @!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        -  /          Z  \                                                	          ;  6	          h	  c
          
                                              5         6  R         4  P         Q  m         n                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                  0  .          `  ^                     "  
$         $  +%         ,%  K&         L&  k'         l'  (         (  )         )  *         *  +         +  -         -  +.         ,.  K/         L/  k0         l0  1         1  2         2  3         3  4         4  6         6  +7         ,7  K8         L8  k9         l9  :         :  ;         ;  <         <  =         =  ?         ?  +@         ,@  KA         LA  kB         lB  C         C  D         D  E         AF  `G         aG  H         H  I         I  J         J  K         K   M         M   N         !N  @O         AO  `P         aP  Q         Q  R         R  S         S  T         T   V         V   W         !W  @X         AX  `Y         aY  Z         Z  [         [  \         \  ]         ]   _         _   `         !`  @a         Aa  `b         ab  c         c  d         d  e         e  f         f   h         h  i         i  k         k  0l         2l  Pm         Rm  pn         rn  o         o  p         p  q         q  r         r  t         t  0u         2u  Pv         Rv  pw         rw  x         x  y         y  z         z  {         {  }         }  0~         2~  P         R  p         r                        Ѓ         ҃                        0         2  P         R  p                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               /!         8!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 s
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-;-x    ;-I-|    I-|-\    |-o-    o--h    -ib-    ib-x-	    x--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               <!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 7	         8	  T
         U
  q         S  o         p                                                                     >  ?          k  l                                                !           L   N!          y!  {"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            (         *  R         T  |         ~                                   $         &  N	         P	  x
         z
                                             "  J         L  t         v                                              F         H  p         r                                              B         D  l         n               !         !  "         $  =%         >%  g&         h&  '         '  (         (  )         )  +         +  9,         :,  c-         d-  .         .  /         /  0         0  2         2  53         63  _4         `4  5         5  6         6  7         7  9         9  1:         2:  [;         \;  <         <  =         =  >         >  @         @  -A         .A  WB         XB  C         C  D         D  E         E  F          G  )H         :H  cI         dI  J         J  K         K  L         L  N         N  5O         6O  _P         `P  Q         Q  R         R  S         S  U         U  1V         2V  [W         \W  X         X  Y         Y  Z         Z  \         \  -]         .]  W^         X^  _         _  `         `  a         a  b          c  )d         *d  Se         Te  }f         ~f  g         g  h          Rm  Fn          rn  fo          o  p          p  q          q  r         ʎ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               3!         F!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-a-     a--    -/;-x    /;-I-|    I-}-\    }-m-    m--h    -Wb-    Wb-?-	    ?--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               .!         8!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         1  M         N  j         k  	         	  
         
           q                                                                     7  &          N  =          e  T          |  k                                    %           J   b!          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       :         <  X         Z  v         x                                              
         
  *         ,  H         J  f         h                                                                    8         :  V         X  t         v                                              
           (         *  F         H  d          f   !         p"  #         #  $         $  %         %  &         &  (         (  ")         $)  @*         B*  ^+         `+  |,         ~,  -         -  .         .  /         /  0         0  2         2  03         23  N4         P4  l5         n5  6         6  7         7  8         8  9         9  ;         ;   <         "<  >=         @=  \>         ^>  z?         |?  @         @  A         A  B         B  C         C  E         E  .F         0F  LG         :H  VI         XI  tJ         vJ  K         K  L         L  M         M  N         N  
P         P  (Q         *Q  FR         HR  dS         fS  T         T  U         U  V         V  W         W  X         X  Z         Z  6[         8[  T\         V\  r]         t]  ^         ^  _         _  `         `  a         a  c         
c  &d         (d  De         Fe  bf                      .  D          j                                             >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               *!         @!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
  j      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-s-     s--    -#5-x    #5-C-|    C-w-\    w--    --h    -Kb-    Kb-x-	    x--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               6!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        =  O          z                                    1            H  +          _  B	          v	  Y
          
  p                                                       4                      ;         <  X         Y  u         v                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         T  A          v  c                                             &  '         '  (         (  
*         *  6+         7+  b,         c,  -         -  .         .  /         /  1         1  >2         ?2  j3         k3  4         4  5         5  6         6  8         8  F9         G9  r:         s:  ;         ;  <         <  =         =  !?         #?  M@         O@  yA         {A  B         B  C         C  D         D  )F         +F  UG         WG  H         H  I         I  J         J  L         L  1M         3M  ]N         _N  O         O  P         P  Q         Q  S         S  9T         ;T  eU         gU  V         V  W         W  X         X  Z         Z  A[         C[  m\         o\  ]         ]  ^         ^  _         _  a         a  Ib         Kb  uc         wc  d         d  e         e  f         f  %h         'h  Qi         Si  }j         j  k         k  l          n  *o         ,o  Vp         Xp  q         q  r         r  s         s  u         u  2v         4v  ^w         `w  x         x  y         y  z         z  |         |  :}         <}  f~         h~                                              B         D  n         p             Ƈ         ȇ                         J         L  v         x             Ύ         Ў             &                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               &!         4!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-[-     [--    -;;-x    ;;-I-|    I-#}-\    #}--    --h    -Qb-    Qb-?-	    ?--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               +!         6!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         1  M         N  j         k  	         	  
         
           v                                              )          *            A  /          X  F          o  ]            t                                    .           T   k!          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       ;         <  Y         Z  w         x                                              
         
  +         ,  I         J  g         h                                                                    9         :  W         X  u         v                                                         )         *  G         H  e          f   !         p"  #         #  $         $  %         %  &         &  (         (  #)         $)  A*         B*  _+         `+  },         ~,  -         -  .         .  /         /  0         0  2         2  13         23  O4         P4  m5         n5  6         6  7         7  8         8  9         9  ;         ;  !<         "<  ?=         @=  ]>         ^>  {?         |?  @         @  A         A  B         B  C         C  E         E  /F         0F  MG         :H  WI         XI  uJ         vJ  K         K  L         L  M         M  N         N  P         P  )Q         *Q  GR         HR  eS         fS  T         T  U         U  V         V  W         W  X         X  Z         Z  7[         8[  U\         V\  s]         t]  ^         ^  _         _  `         `  a         a  	c         
c  'd         (d  Ee         Fe  cf            4          Z  p                      ҏ              $                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               &!         <!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
  i      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-m-     m--    -/5-x    /5-C-|    C-w-\    w--    --h    -Qb-    Qb-x-	    x--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               4!         @!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        =  N          z                                    1            H  *          _  A	          v	  X
          
  o                                                       9           (         )  E         F  b         c                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                  B  /          d  Q            s         &  '         '  (         (  	*         *  5+         7+  a,         c,  -         -  .         .  /         /  1         1  =2         ?2  i3         k3  4         4  5         5  6         6  8         8  E9         G9  q:         s:  ;         ;  <         <  =         =  "?         #?  N@         O@  zA         {A  B         B  C         C  D         D  *F         +F  VG         WG  H         H  I         I  J         J  L         L  2M         3M  ^N         _N  O         O  P         P  Q         Q  S         S  :T         ;T  fU         gU  V         V  W         W  X         X  Z         Z  B[         C[  n\         o\  ]         ]  ^         ^  _         _  a         a  Jb         Kb  vc         wc  d         d  e         e  f         f  &h         'h  Ri         Si  ~j         j  k         k  l          n  +o         ,o  Wp         Xp  q         q  r         r  s         s  u         u  3v         4v  _w         `w  x         x  y         y  z         z  |         |  ;}         <}  g~         h~                                              C         D  o         p             Ǉ         ȇ                         K         L  w         x             ώ         Ў             '                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               *!         8!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,-y-     y--    -;-    ;-J-    J-}-d    }--    --p    -ib-    ib-	-
    	--
    --t    --׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               =!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             "         #  E         F  h         i                                               
  "          G  _                                                          H  I          u  v                                   )!         *!  F"         G"  c#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                    :  .          W  K          t  h         %  &         &  '         '  )         
)  6*         8*  d+         f+  ,         ,  -         -  .         .  0         0  J1         L1  x2         z2  3         3  4         4  6         6  07         27  ^8         `8  9         9  :         :  ;         ;  =         =  D>         F>  r?         t?  @         @  A         A  B         B  *D         ,D  XE         ZE  F         F  G         G  H         H  J         J  >K         @K  lL         nL  M         M  N         N  O         O  %Q         &Q  SR         TR  S         S  T         T  U         U  W         W  9X         :X  gY         hY  Z         Z  [         [  \         \  ^          ^  M_         N_  {`         |`  a         a  b         b  d         d  3e         4e  af         bf  g         g  h         h  i         i  k         k  Gl         Hl  um         vm  n         n  o         o  p          q  ,r         .r  Zs         \s  t         t  u         u  v         v  x         x  @y         By  nz         pz  {         {  |         |  }         }  &         (  T         V                        ރ                      :         <  h         j             ĉ         Ɖ                       "  N         P  |         ~                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               3!         J!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    d
         e
             7      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     P
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-U-     U--    -;-x    ;-I-|    I-|-\    |-,-    ,--h    -Wb-    Wb--	    --h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               '!         8!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 7	         8	  T
         U
  q         S  o         p                                                                     >  @          k  m                                                            L   M!          y!  z"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            -         .  [         \                                              A         B  o	         p	  
         
                        '         (  U         V                                              ;         <  i         j                                   !         "  O         P  }         ~                        !         !  5"         6"  c#         %  &         &  '         '  %)         &)  S*         T*  +         +  ,         ,  -         -  /         /  90         :0  g1         h1  2         2  3         3  4         4  6          6  M7         N7  {8         |8  9         9  :         :  <         <  3=         4=  a>         b>  ?         ?  @         @  A         A  C         C  GD         HD  uE         vE  F         F  G         G  H         TI  J         J  K          d  e          e  g          0g  Dh          ph  i          i  j           ?         A  m         o             ɐ         ː             %         '  S         U                        ݗ         ߗ             9         ;  g         i             Ý         ŝ                      !  M         O  {         }             פ         ٤             3         5  a         c                                                      >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               -!         @!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   3      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     --    -;-    ;-J-    J-}-d    }--    --p    -ib-    ib-	-
    	--
    --t    --׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ?!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             #         $  G         H  k         l                                                 &          L  c                                                 "          M  O          z  |                                   .!         /!  K"         L"  h#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                               	          
  
         %  &         &  '         '  	)         
)  7*         8*  e+         f+  ,         ,  -         -  .         .  0         0  K1         L1  y2         z2  3         3  4         4  6         6  17         27  _8         `8  9         9  :         :  ;         ;  =         =  E>         F>  s?         t?  @         @  A         A  B         B  +D         ,D  YE         ZE  F         F  G         G  H         H  J         J  ?K         @K  mL         nL  M         M  N         N  O         O  $Q         &Q  RR         TR  S         S  T         T  U         U  
W         W  8X         :X  fY         hY  Z         Z  [         [  \         \  ^          ^  L_         N_  z`         |`  a         a  b         b  d         d  2e         4e  `f         bf  g         g  h         h  i         i  k         k  Fl         Hl  tm         vm  n         n  o         o  p          q  -r         .r  [s         \s  t         t  u         u  v         v  x         x  Ay         By  oz         pz  {         {  |         |  }         }  '         (  U         V                        ߃                      ;         <  i         j             ŉ         Ɖ             !         "  O         P  }         ~                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               3!         J!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    d
         e
             6      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     Q
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-U-     U--    -;-x    ;-I-|    I-|-\    |-,-    ,--h    -Qb-    Qb--	    --h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               &!         6!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 7	         8	  T
         U
  q         S  o         p                                                                     >  ?          k  l                                                !           L   N!          y!  {"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ,         .  Z         \                                              @         B  n	         p	  
         
                        &         (  T         V                                              :         <  h         j                                             "  N         P  |         ~                        !         !  4"         6"  b#         %  &         &  '         '  $)         &)  R*         T*  +         +  ,         ,  -         -  
/         /  80         :0  f1         h1  2         2  3         3  4         4  6          6  L7         N7  z8         |8  9         9  :         :  <         <  2=         4=  `>         b>  ?         ?  @         @  A         A  C         C  FD         HD  tE         vE  F         F  G         G  H         TI  J         J  K          j  l          0l  Dm          n  o          o  q          0q  Dr           @         A  n         o             ʐ         ː             &         '  T         U                        ޗ         ߗ             :         ;  h         i             ĝ         ŝ                       !  N         O  |         }             ؤ         ٤             4         5  b         c                                                      >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               )!         <!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   4      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    <     -,-p     ,-y-     y--    -M5-    M5-C-    C-5w-t    5w-3-     3-;-X    ;--    -ob-p	    ob--x
    --
    -a-    a-ճ-\    ճ--<                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ;!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        -  .          Z  [                                                
          ;  7	          h	  d
          
                                              5         6  R         4  P         Q  m         n                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        D%  n&         JL  uM         vM  N         N  O         O  P         P  %R         &R  QS         RS  }T         ~T  U         U  V         V  X         X  -Y         .Y  YZ         ZZ  [         [  \         \  ]         ]  	_         
_  5`         6`  aa         ba  b         b  c         c  d         d  f         f  =g         >g  ih         jh  i         i  j         j  k         k  m         m  En         Fn  qo         ro  p         p  q         q  r         r  !t         "t  Mu         Nu  yv         zv  w         w  x         x  y         y  ){         *{  U|         V|  }         }  ~         ~                        1         2  ]         ^                                              9         :  e         f                                              A         B  m         n             ē         Ɠ                        H         J  t         v             ̚         Κ             $          &  )          R  U          ~                        ֡  ٢           ,         .  X         Z                        ܨ         ި           
  4         6  `         b                                              <         >  h         j                                b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               <!         L!  1         1  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    d
         e
             7                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    d
         e
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            4     -,-`     ,-m-     m--    -;-    ;-kJ-    kJ-}-d    }-	-    	--p    -ob-    ob--
    --
    -ϰ-t    ϰ-'׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               6!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        9         :  V         W  s         t                                                I  J          v  w                                    /          Z  l                                             !         !  ("         )"  E#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       0  7          `  g                                 #  $         $  %         %  &         '  #(         *(  G)         N)  k*         r*  +         +  ,         ,  -         -  .         /  0         &0  C1         J1  g2         n2  3         3  4         4  5         5  6         6  8         "8  ?9         F9  c:         j:  ;         ;  <         <  =         =  >         >  @         @  ;A         BA  _B         fB  C         C  D         D  E         E  F         F  H         H  6I         ;I  WJ         \J  xK         }K  L         L  M         M  N         N  P         
P  ,Q         .Q  PR         RR  tS         vS  T         T  U         U  V         V  X         X  &Y         (Y  HZ         JZ  j[         l[  \         \  ]         ]  ^         ^  _         _  a         a  6b         8b  Xc         Zc  xd         zd  e         e  f         f  g         g  h         h  j         j  8k         :k  Xl         Zl  xm         zm  n         n  o         o  p         p  q         q  s         s  :t         ?t  [u         `u  |v         v  w         w  x         x  y         y   {         {  !|         &|  B}         G}  c~         h~                        Ɓ         ˁ                        )         .  J         O  k         p                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               @!         J!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    d
         e
             .      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     7
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    <     -,-p     ,--     --    -M5-    M5-C-    C-5w-t    5w-4-     4-;-X    ;--    -ob-p	    ob--x
    --
    -a-    a-ճ-\    ճ--<                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               =!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        -  /          Z  \                                                	          ;  6	          h	  c
          
                                              5         6  R         4  P         Q  m         n                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        D%  o&         JL  tM         vM  N         N  O         O  P         P  $R         &R  PS         RS  |T         ~T  U         U  V         V   X         X  ,Y         .Y  XZ         ZZ  [         [  \         \  ]         ]  _         
_  4`         6`  `a         ba  b         b  c         c  d         d  f         f  <g         >g  hh         jh  i         i  j         j  k         k  m         m  Dn         Fn  po         ro  p         p  q         q  r         r   t         "t  Lu         Nu  xv         zv  w         w  x         x  y         y  ({         *{  T|         V|  }         }  ~         ~                        0         2  \         ^                                              8         :  d         f                                              @         B  l         n             œ         Ɠ                        I         J  u         v             ͚         Κ             %         &  Q         R  }         ~             ա         ֡                        .  1          Z  ]                                 ި  	         
  5         6  a         b                                              =         >  i         j                                b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               >!         L!  1         1  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    d
         e
             6                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    d
         e
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            4     -,-`     ,-y-     y--    -;-    ;-kJ-    kJ-}-d    }-	-    	--p    -ob-    ob--
    --
    -ϰ-t    ϰ-'׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               :!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        9         :  V         W  s         t                                                I  K          v  x                                    .          Z  k                                             !         !  ,"         -"  J#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                        '          P  W	          	  
          
           #  $         $  %         %  &         '  "(         *(  F)         N)  j*         r*  +         +  ,         ,  -         -  .         /  0         &0  B1         J1  f2         n2  3         3  4         4  5         5  6         6  8         "8  >9         F9  b:         j:  ;         ;  <         <  =         =  >         >  @         @  :A         BA  ^B         fB  C         C  D         D  E         E  F         F  H         H  7I         ;I  XJ         \J  yK         }K  L         L  M         M  N         N  	P         
P  -Q         .Q  QR         RR  uS         vS  T         T  U         U  V         V  X         X  'Y         (Y  IZ         JZ  k[         l[  \         \  ]         ]  ^         ^  _         _  a         a  7b         8b  Yc         Zc  yd         zd  e         e  f         f  g         g  h         h  j         j  9k         :k  Yl         Zl  ym         zm  n         n  o         o  p         p  q         q  s         s  ;t         ?t  \u         `u  }v         v  w         w  x         x  y         y  {         {  "|         &|  C}         G}  d~         h~                        ǁ         ˁ             	           *         .  K         O  l         p                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               >!         J!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    d
         e
             /      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     8
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-s-     s--    -;-x    ;-J-|    J-}-\    }-96-    96--h    -b-    b-n-	    n--h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               8!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             "         #  E         F  h         i                                               
  "          G  _                                                          H  I          u  v                                   )!         *!  F"         G"  c#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            (         *  R         T  |         ~                                   $         &  N	         P	  x
         z
                                             "  J         L  t         v                                              F         H  p         r                                              B         D  l         n               !         !  "          =  >          >  @          0@  CA          lA  B          B  C         K  L         L  N         N  1O         2O  [P         \P  Q         Q  R         R  S         S  U         U  -V         .V  WW         XW  X         X  Y         Y  Z         Z  [          \  )]         *]  S^         T^  }_         ~_  `         `  a         a  b         b  %d         &d  Oe         Pe  yf         zf  g         g  h         h  i         i  !k         "k  Kl         Ll  um         vm  n         n  o           6         8  `         b                        ޚ                    
  2         4  \         ^                        ڡ         ܡ             .         0  X         Z                        ֨         ب              *         ,  T         V  ~                      ү         ԯ             &         (  P         R  z         |             ζ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               Q!         R!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     <
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                 w        Yܯ--0     --     -c-     c-B-P    B-c-    c--    -'׳-p                             b      *      7   I      operation.0     sync             :            2            b         reason  ExchangeBarrier value   1       invocations     1                                     b      *      7   I      operation.0     sync             :            2            b         reason  ExchangeBarrier value   3       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   5       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   2       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-s-     s--    -;-x    ;-J-|    J-}-\    }-96-    96--h    -b-    b-n-	    n--h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               :!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             #         $  G         H  k         l                                                 &          L  c                                                 "          M  O          z  |                                   .!         /!  K"         L"  h#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            )         *  S         T  }         ~                                   %         &  O	         P	  y
         z
                                   !         "  K         L  u         v                                              G         H  q         r                                              C         D  m         n               !         !  "          C  D           E  3F          \F  oG          G  H          H  I         K  L         L  N         N  0O         2O  ZP         \P  Q         Q  R         R  S         S  U         U  ,V         .V  VW         XW  X         X  Y         Y  Z         Z  [          \  (]         *]  R^         T^  |_         ~_  `         `  a         a  b         b  $d         &d  Ne         Pe  xf         zf  g         g  h         h  i         i   k         "k  Jl         Ll  tm         vm  n         n  o           7         8  a         b                        ߚ           	         
  3         4  ]         ^                        ۡ         ܡ             /         0  Y         Z                        ר         ب             +         ,  U         V                        ӯ         ԯ             '         (  Q         R  {         |             ϶                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               U!         V!   2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     ;
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                 w        Yܯ--0     --     -c-     c-B-P    B-c-    c--    -'׳-p                             b      *      7   I      operation.0     sync             :            2            b         reason  ExchangeBarrier value   1       invocations     1                                     b      *      7   I      operation.0     sync             :            2            b         reason  ExchangeBarrier value   3       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   5       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   2       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,-=-     =--    -_5-    _5-C-    C-Gw-l    Gw-90-    90--x    -ub-	    ub--
    --
    -[-|    [-ճ-    ճ-!׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     f            !          J!  1          1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         +  (          V  S            ~                                    
          5  9	          d	  h
          
                                               =         >  \         H  f         g                                                                 b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |        T$  v%         x%  &          &  '          '  (          ")  7*          e*  z+          +  ,         vL  M         M  N         N  O         O  Q         Q  )R         *R  MS         NS  qT         rT  U         U  V         V  W         W  Y         Y  %Z         &Z  I[         J[  m\         n\  ]         ]  ^         ^  _         _  `         `  !b         "b  Ec         Fc  id         jd  e         e  f         f  g         g  h         h  j         j  Ak         Bk  el         fl  m         m  n         n  o         o  p         p  r         r  =s         >s  at         bt  u         u  v         v  w         w  x         x  z         z  9{         :{  ]|         ^|  }         }  ~         ~                                   5         6  Y         Z  }         ~             Ň         Ƈ                        1         2  U         V  y         z                                            
  ,         .  P         R  t         v                                              (         *  L         N  p         r                        ܠ         ޠ              $         &  H         J  l         n                        ب         ڨ                       "  D         F  h         j                        ԰                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               K!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    f
         g
             9      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    f
         g
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -{,-h     {,-1-     1--    -;-    ;-_J-    _J-}-l    }-a-    a--P    --    -b-h	    b--p
    --
    --    --L                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     i            !          >!  1          1  v=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        =         >  \         ]  {         |                                                C  @          n  k                                    %          J  `                                                        !  ""         #"  E#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        $  %         %  &         &  (         (  ')         ()  M*         N*  s+         t+  ,         ,  -         -  .         .  0         0  11         21  W2         X2  }3         ~3  4         4  5         5  6         6  8         8  ;9         <9  a:         b:  ;         ;  <         <  =         =  >         >  @          @  EA         FA  kB         lB  C         C  D         D  E         E  G          G  G          *H  %I          PI  KJ          vJ  qK          K  L         L  M         M  O         O  2P         4P  XQ         ZQ  ~R         R  S         S  T         T  U         U  W         W  <X         >X  bY         dY  Z         Z  [         [  \         \  ]         ]   _         "_  F`         H`  la         na  b         b  c         c  d         d  f         
f  0g         2g  Xh         Zh  i         i  j         j  k         k  l         l   n         "n  Ho         Jo  np         pp  q         q  r         r  s         s  u         u  ,v         .v  Rw         Tw  xx         zx  y         y  z         z  {         {  }         }  6~         8~  \         ^                        ΂         Ђ                        @         B  f         h                        ؊         ڊ                                b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               U!         V!   2         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    f
         g
             8                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     3
                   
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            8     -,-h     ,-=-     =--    -_5-    _5-C-    C-Gw-l    Gw-90-    90--x    -ib-	    ib--
    --
    -[-|    [-ճ-    ճ-!׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     e            
!          J!  1          1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         +  )          V  T                                                	          5  8	          d	  g
          
                                               =         >  \         H  f         g                                                                 b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |        T$  w%         x%  &          ,   .          ..  C/          q/  0          0  1          1  3         vL  M         M  N         N  O         O  Q         Q  (R         *R  LS         NS  pT         rT  U         U  V         V  W         W   Y         Y  $Z         &Z  H[         J[  l\         n\  ]         ]  ^         ^  _         _  `         `   b         "b  Dc         Fc  hd         jd  e         e  f         f  g         g  h         h  j         j  @k         Bk  dl         fl  m         m  n         n  o         o  p         p  r         r  <s         >s  `t         bt  u         u  v         v  w         w  x         x  z         z  8{         :{  \|         ^|  }         }  ~         ~                                   4         6  X         Z  |         ~             ć         Ƈ                        0         2  T         V  x         z                                   	         
  -         .  Q         R  u         v                                              )         *  M         N  q         r                        ݠ         ޠ             %         &  I         J  m         n                        ٨         ڨ             !         "  E         F  i         j                        հ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               G!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    f
         g
             8      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    f
         g
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -u,-h     u,-+-     +--    -;-    ;-eJ-    eJ-}-l    }-a-    a--P    --    -b-h	    b--p
    --
    -ð-    ð--L                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     h            !          >!  1          1  u=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        =         >  \         ]  {         |                                                C  A          n  l                                    $          J  _                                             !         !  &"         '"  J#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        $  %         %  &         &   (         (  &)         ()  L*         N*  r+         t+  ,         ,  -         -  .         .  
0         0  01         21  V2         X2  |3         ~3  4         4  5         5  6         6  8         8  :9         <9  `:         b:  ;         ;  <         <  =         =  >         >  @          @  DA         FA  jB         lB  C         C  D         D  E         E  G         G  )H         *H  OI         PI  uJ         vJ  K         K  L         L  M         M  O         O  3P         4P  YQ         ZQ  R         R  S         S  T         T  U         U  W         W  =X         >X  cY         dY  Z         Z  [         [  \         \  ]         ]  !_         "_  G`         H`  ma         na  b         b  c         c  d         d  	f         
f  1g         2g  Yh         Zh  i         i  j         j  k         k  l         l  !n         "n  Io          Jo  Ep          pp  kq          q  r          r  s          s  t         u  -v         .v  Sw         Tw  yx         zx  y         y  z         z  {         {  }         }  7~         8~  ]         ^                        ς         Ђ                        A         B  g         h                        ي         ڊ                                b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               U!         V!   2         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    f
         g
             9                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     4
                   
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,-O-     O--    -;-x    ;-MJ-|    MJ-}-\    }-c3-    c3--h    -b-    b-m-	    m--h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     h            !          P!  1          1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        =         >  \         ]  {         |            r  ~                                  #  /          ^  j                                                          E  B                                              "         "  1#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                      '  G         N  n         u                                   
           1	         8	  X
         _
                                                       "  B         I  i         p                                              ,         3  S         Z  z                                                       =          D   d!         k!  "          /  0          1  2          J2  V3          3  4          4  5         K  L         L  N         N  5O         >O  cP         lP  Q         Q  R         R  S         S  U         $U  IV         RV  wW         W  X         X  Y         Y  Z         [  '\         0\  S]         \]  ^         ^  _         _  `         `  b         b  /c         8c  Yd         bd  e         e  f         f  g         g  i         
i  +j         4j  Uk         ^k  l         l  m         m  n         ȑ                        0         4  T         X  x         |                      ę                        ,         0  P         T  t         x                                              (         ,  L         P  p         t                        ܪ                       $         (  H         L  l         p                        ز         ܲ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               R!         V!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     0
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            8     -,-h     ,-a-     a--    -4-    4-C-    C-v-l    v---    ---x    --b-	    -b--
    --
    -m-|    m-ճ-    ճ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               1!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        ;  Q          v                                    '            <  +          Q  @	          f	  U
          {
  j                                                       >                                 6         7  U         V  t                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         r  l                                                          V"  {#         }#  $         H  I         I  J         J  K         K  M         M  EN         GN  mO         oO  P         P  Q         Q  R         R  T         T  5U         7U  ]V         _V  W         W  X         X  Y         Y  Z         Z  %\         '\  M]         O]  u^         w^  _         _  `         `  a         a  c         c  =d         ?d  ee         ge  f         f  g         g  h         h  j         j  -k         /k  Ul         Wl  }m         m  n         n  o         o  p         p  r         r  Es         Gs  mt         ot  u         u  v         v  w         w  y         y  5z         7z  ]{         _{  |         |  }         }  ~         ~             %         '  M         O  u         w             Ņ         ǅ                        =         ?  e         g                        ݍ         ߍ             ,         .  S         U  {         }             ˔         ͔                        C         E  k         m                                              3         5  [         ]                        ӣ         գ             #         %  K         M  s         u             ê         Ū                        ;                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     i            !          D!  1          1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    f
         g
             8      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    h
         i
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            0     -,-X     ,-O-     O--    -;-x    ;-MJ-|    MJ-}-\    }-c3-    c3--h    -b-    b-m-	    m--h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     i            !          P!  1          1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        =         >  \         ]  {         |            r  }                                  #  .          ^  i                                                          E  C                                              "         "  1#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            !         '  H         N  o         u                                              2	         8	  Y
         _
                                                       "  C         I  j         p                                              -         3  T         Z  {                                                       >          D   e!         k!  "          5  7          27  >8          l8  x9          9  :          :  ;         K  L         L  N         N  4O         >O  bP         lP  Q         Q  R         R  S         S  U         $U  HV         RV  vW         W  X         X  Y         Y  Z         [  &\         0\  R]         \]  ~^         ^  _         _  `         `  b         b  .c         8c  Xd         bd  e         e  f         f  g         g   i         
i  *j         4j  Tk         ^k  ~l         l  m         m  n         ȑ                        1         4  U         X  y         |                      ę             	           -         0  Q         T  u         x                                              )         ,  M         P  q         t                        ݪ                      %         (  I         L  m         p                        ٲ         ܲ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               P!         V!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     /
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            8     -,-h     ,-[-     [--    -4-    4-C-    C-v-l    v---    ---x    -'b-	    'b--
    --
    -m-|    m-ճ-    ճ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               /!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        ;  P          v                                    '            <  *          Q  ?	          f	  T
          {
  i                                                        C                      !         "  @         A  _         `  ~                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         +  %          P  J          u  o                                 V"  |#         }#  $         H  I         I  J         J  K         K  M         M  FN         GN  nO         oO  P         P  Q         Q  R         R  T         T  6U         7U  ^V         _V  W         W  X         X  Y         Y  Z         Z  &\         '\  N]         O]  v^         w^  _         _  `         `  a         a  c         c  >d         ?d  fe         ge  f         f  g         g  h         h  j         j  .k         /k  Vl         Wl  ~m         m  n         n  o         o  p         p  r         r  Fs         Gs  nt         ot  u         u  v         v  w         w  y         y  6z         7z  ^{         _{  |         |  }         }  ~         ~             &         '  N         O  v         w             ƅ         ǅ                        >         ?  f         g                        ލ         ߍ             -         .  T         U  |         }             ̔         ͔                        D         E  l         m                                              4         5  \         ]                        ԣ         գ             $         %  L         M  t         u             Ī         Ū                        <                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     h            !          @!  1          1  x=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    f
         g
             9      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    i
         j
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            0     -,-X     ,-y-     y--    -k5-x    k5-D-|    D-Sw-\    Sw-(-    (--h    -Kb-    Kb-m-	    m--h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ;!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         +  (          V  S            ~                                    
          5  9	          d	  h
          
                                               =         >  \         H  f         g                                                                 b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                     %  C         J  h         o                                              !	         (	  F
         M
  k         r                                              $         +  I         P  n         u                                            	  '         .  L         S  q         x                                                *!         1!  O"         I  J         J  K         K  L         M  !N         (N  FO         MO  kP         rP  Q         Q  R         R  S         S  T         U  $V         +V  IW         PW  nX         uX  Y         Y  Z         Z  [         [  ]         	]  '^         .^  L_         S_  q`         x`  a         a  b         b  c         c  	e         e  /f         6f  Ug         \g  {h         h  i         i  j         j  k         k  m         m  ;n         Dn  co         lo  p         p  q         q  r         r  t         t  +u         4u  Sv         \v  {w         w  x         x  y         y  z         z  |         $|  C}         L}  k~         t~                      Ā                        3         <  [         d                        ӈ         ܈             #         ,  K         T  s         |                        ¤                                    @  <                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     l            !          T!  1          1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     0
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-m-     m--    -;-x    ;-eJ-|    eJ-}-\    }-.-    .--h    -Eb-    Eb-n-	    n--h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               6!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        =         >  \         ]  {         |                                                C  @          n  k                                    %          J  `                                                        !  ""         #"  E#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            (         *  R         T  |         ~                                   $         &  N	         P	  x
         z
                                             "  J         L  t         v                                              F         H  p         r                                              B         D  l         n               !          `;  j<          <  =          =  >          >  ?           @  *A         @G  iH         jH  I         I  J         J  K         K  "M         $M  LN         NN  vO         xO  P         P  Q         Q  R         R  T          T  HU         JU  rV         tV  W         W  X         X  Y         Y  [         [  D\         F\  n]         p]  ^         ^  _         _  `         `  b         b  @c         Bc  jd         ld  e         e  f         f  g         g  i         i  <j         >j  fk         hk  l         l  m           ͏         Ώ             !         "  K         L  u         v             ɖ         ʖ                        G         H  q         r             ŝ         Ɲ                        C         D  m         n                      ¤                        ?         @  i         j                                              ;                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     k            !          N!  1          1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     O
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     --    -k5-x    k5-D-|    D-Sw-\    Sw-w.-    w.--h    -Eb-    Eb-m-	    m--h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               =!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         +  )          V  T                                                	          5  8	          d	  g
          
                                               =         >  \         H  f         g                                                                 b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                     %  D         J  i         o                                              "	         (	  G
         M
  l         r                                               %         +  J         P  o         u                                            	  (         .  M         S  r         x                                                +!         1!  P"         I  J         J  K         K  L         M  "N         (N  GO         MO  lP         rP  Q         Q  R         R  S         S   U         U  %V         +V  JW         PW  oX         uX  Y         Y  Z         Z  [         [  ]         	]  (^         .^  M_         S_  r`         x`  a         a  b         b  c         c  
e         e  0f         6f  Vg         \g  |h         h  i         i  j         j  k         k  m         m  :n         Dn  bo         lo  p         p  q         q  r         r  t         t  *u         4u  Rv         \v  zw         w  x         x  y         y  z         z  |         $|  B}         L}  j~         t~                      Ā             
           2         <  Z         d                        ҈         ܈             "         ,  J         T  r         |            j  f                                                                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     k            !          P!  1          1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     /
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-m-     m--    -;-x    ;-kJ-|    kJ-}-\    }-.-    .--h    -Eb-    Eb-n-	    n--h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               6!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        =         >  \         ]  {         |                                                C  A          n  l                                    $          J  _                                             !         !  &"         '"  J#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            )         *  S         T  }         ~                                   %         &  O	         P	  y
         z
                                   !         "  K         L  u         v                                              G         H  q         r                                              C         D  m         n               !          PA  ZB          B  C          C  D          D  E          F  G         @G  hH         jH  I         I  J         J  K         K  #M         $M  MN         NN  wO         xO  P         P  Q         Q  R         R  T          T  IU         JU  sV         tV  W         W  X         X  Y         Y  [         [  E\         F\  o]         p]  ^         ^  _         _  `         `  b         b  Ac         Bc  kd         ld  e         e  f         f  g         g  i         i  =j         >j  gk         hk  l         l  m           ̏         Ώ                       "  J         L  t         v             Ȗ         ʖ                        F         H  p         r             ĝ         Ɲ                        B         D  l         n                      ¤                        >         @  h         j                                              :                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     j            !          N!  1          1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     N
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-;-x    ;-I-|    I-|-\    |-1-    1-w-h    w-b-    b-'-	    '--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               C!         X!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         '  E         F  d         e  	         	  
         
                                              
           )          *            ?            T  /          i  D          ~  Y                                  	             D   P!          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            .         0  ^         `                                               N         P  ~	         	  
         
                        >         @  n         p                                    .         0  ^         `                                               N         P  ~                                              >!         @!  n"         p"  #         #  $         $  %          &  .'         0'  ^(         `(  )         )  *         *  +         +  -          -  N.         P.  ~/         /  0         0  1         1  3         3  >4         @4  n5         p5  6         6  7         7  8          9  .:         0:  ^;         `;  <         <  =         =  >         >  @          @  NA         PA  ~B         B  C         C  D         D  F         F  >G         @G  nH         pH  I         I  J          l  m          m  n          o   p          Sp  cq          q  r         d                      Ĕ             "         $  R         T                                              B         D  r         t             ҡ         ԡ             2         4  b         d             §         ħ             "         $  R         T                                              B         D  r                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     \                       6!  1          1  ]=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   Z      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-s-     s--    -4-x    4-C-|    C-v-\    v--    --h    -ob-    ob-n-	    n--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               8!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        ;  Q          v                                    '            <  +          Q  @	          f	  U
          {
  j                                                       >                                 6         7  U         V  t                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            *         ,  V         X                                              2         4  ^	         `	  
         
                                   :         <  f         h                                              B         D  n         p                                               J         L  v         x                          !         !  &#          F  G          H  I         HI  rJ         tJ  K         K  L         L  M         M  "O         $O  NP         PP  zQ         |Q  R         R  S         S  T          U  *V         ,V  VW         XW  X         X  Y         Y  Z         Z  \         \  2]         4]  ^^         `^  _         _  `         `  a         a  c         c  :d         <d  fe         he  f         f  g         g  h         h  j         j  Bk         Dk  nl         pl  m         m  n         n  o         o  q          q  Jr         Lr  vs         xs  t         t  u         u  v         v  &x         (x  Ry         Ty  ~z         z  {         {  |         |  ~         ~  .         0  Z         \                        ރ           
           6         8  b         d                                              >         @  j         l                        ؐ              /                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               <!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     O
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-;-x    ;-I-|    I-|-\    |-1-    1-}-h    }-b-    b-'-	    '--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               A!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         '  E         F  d         e  	         	  
         
                                                         .          /  	          D            Y  3          n  H            ]                                               I   T!          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            /         0  _         `                                               O         P  	         	  
         
                        ?         @  o         p                                    /         0  _         `                                               O         P                                                ?!         @!  o"         p"  #         #  $         $  %          &  /'         0'  _(         `(  )         )  *         *  +         +  -          -  O.         P.  /         /  0         0  1         1  3         3  ?4         @4  o5         p5  6         6  7         7  8          9  /:         0:  _;         `;  <         <  =         =  >         >  @          @  OA         PA  B         B  C         C  D         D  F         F  ?G         @G  oH         pH  I         I  J          r  s          t  ,u          _u  ov          v  w          w  x         d             Ô         Ĕ             #         $  S         T                                              C         D  s         t             ӡ         ԡ             3         4  c         d             ç         ħ             #         $  S         T                                              C         D  s                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     ]                       :!  1          1  d=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   Y      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-s-     s--    -5-x    5-C-|    C-v-\    v--    --h    -ib-    ib-n-	    n--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               8!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        ;  P          v                                    '            <  *          Q  ?	          f	  T
          {
  i                                                        C                      !         "  @         A  _         `  ~                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            +         ,  W         X                                              3         4  _	         `	  
         
                                   ;         <  g         h                                              C         D  o         p                                               K         L  w         x                          !         !  '#         HI  sJ         tJ  K         K  L         L  M         M  #O         $O  OP         PP  {Q         |Q  R         R  S         S  T          U  +V         ,V  WW         XW  X         X  Y         Y  Z         Z  \         \  3]         4]  _^         `^  _         _  `         `  a         a  c         c  ;d         <d  ge         he  f         f  g         g  h         h  j         j  Ck         Dk  ol         pl  m         m  n         n  o         o  q          q  Kr         Lr  ws         xs  t         t  u         u  v         v  'x         (x  Sy         Ty  z         z  {         {  |         |  ~         ~  /         0  [         \                        ߃                      7         8  c         d                                              ?         @  k         l            X  o                      ؕ              /          X  o                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               8!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     N
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-y-     y--    -;-x    ;-YJ-|    YJ-}-\    }--    --h    -3b-    3b---	    ---h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               7!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        =         >  \         ]  {         |            r  ~                                  #  /          ^  j                                                          E  B                                              "         "  1#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ,         .  Z         \                                              @         B  n	         p	  
         
                        &         (  T         V                                              :         <  h         j                                             "  N         P  |         ~                        !         !  4"         6"  b#         d#  $         $  %         %  &         &  (         (  H)         J)  v*         x*  +         +  ,         ,   .         .  ./         0/  \0         ^0  1         1  2         2  3         3  5         5  B6         D6  p7         r7  8         8  9         9  :         :  (<         *<  V=         X=  >         >  ?         ?  @         @  B         B  <C         >C  jD         lD  E         E  F          S  T          *U  8V          hV  vW          W  X          X  Y         l  m         m  o          o  Mp         Np  {q         |q  r         r  s         s  u         u  3v         4v  aw         bw  x         x  y         y  z         z  |         |  G}         H}  u~         v~             р         Ҁ              -         .  [         \                                              A         B  o         p             ˍ         ̍             '                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               !         *!  1         1  u=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   Z      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-;-x    ;-I-|    I-}-\    }-g-    g--h    -Wb-    Wb-x-	    x--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               @!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 3	         4	  R
         S
  q         ]  {         |                                                         '  !          V  P                                                          9  6           d   a!          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            -         .  [         \                                              A         B  o	         p	  
         
                        '         (  U         V                                              ;         <  i         j                                   !         "  O         P  }         ~                        !         !  5"         6"  c#         d#  $         $  %         %  &         &  (         (  H)         J)  v*         x*  +         +  ,         ,   .         .  ./         0/  \0         ^0  1         1  2         2  3         3  5         5  B6         D6  p7         r7  8         8  9         9  :         :  (<         *<  V=         X=  >         >  ?         ?  @         @  B         B  <C         >C  jD         lD  E         E  F         F  G         G  "I         $I  PJ         RJ  ~K         K  L         L  M         M  O         
O  6P         8P  dQ         fQ  R         R  S         S  T         T  V         V  JW         LW  xX         zX  Y         Y  Z         Z  \         \  0]         2]  ^^         `^  _         _  `         `  a         a  c         c  Dd         Fd  re         te  f         f  g         g  h          ft  Tu          u  xv          v  w          w  x          x  y         ʐ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               -!         >!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-s-     s--    -;-x    ;-YJ-|    YJ-}-\    }--    --h    -9b-    9b---	    ---h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               5!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        =         >  \         ]  {         |            r  }                                  #  .          ^  i                                                          E  C                                              "         "  1#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            -         .  [         \                                              A         B  o	         p	  
         
                        '         (  U         V                                              ;         <  i         j                                   !         "  O         P  }         ~                        !         !  5"         6"  c#         d#  $         $  %         %  &         &  (         (  I)         J)  w*         x*  +         +  ,         ,  .         .  //         0/  ]0         ^0  1         1  2         2  3         3  5         5  C6         D6  q7         r7  8         8  9         9  :         :  )<         *<  W=         X=  >         >  ?         ?  @         @  B         B  =C         >C  kD         lD  E         E  F          "Z  0[          `[  n\          \  ]          ]  ^          _  (`         l  m         m  o          o  Lp         Np  zq         |q  r         r  s         s  u         u  2v         4v  `w         bw  x         x  y         y  z         z  |         |  F}         H}  t~         v~             Ѐ         Ҁ              ,         .  Z         \                                              @         B  n         p             ʍ         ̍             &                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               !         .!  1         1  }=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   Y      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     --    -;-x    ;-I-|    I-}-\    }-g-    g--h    -]b-    ]b-x-	    x--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               <!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 3	         4	  R
         S
  q         ]  {         |                                                         '             V  O            ~                                              9  7           d   b!          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ,         .  Z         \                                              @         B  n	         p	  
         
                        &         (  T         V                                              :         <  h         j                                             "  N         P  |         ~                        !         !  4"         6"  b#         d#  $         $  %         %  &         &  (         (  I)         J)  w*         x*  +         +  ,         ,  .         .  //         0/  ]0         ^0  1         1  2         2  3         3  5         5  C6         D6  q7         r7  8         8  9         9  :         :  )<         *<  W=         X=  >         >  ?         ?  @         @  B         B  =C         >C  kD         lD  E         E  F         F  G         G  #I         $I  QJ         RJ  K         K  L         L  M         M  	O         
O  7P         8P  eQ         fQ  R         R  S         S  T         T  V         V  KW         LW  yX         zX  Y         Y  Z         Z  \         \  1]         2]  _^         `^  _         _  `         `  a         a  c         c  Ed         Fd  se         te  f         f  g         g  h          z  {          >{  ,|          b|  P}          }  t~          ~           ʐ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               /!         B!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-s-     s--    -;-x    ;-I-|    I-}-\    }-k
-    k
--P    -ib-    ib-c-	    c-g-D
    g-3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               4!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         '  E         F  d         e  	         	  
         
                                              
           )          *            ?            T  /          i  D          ~  Y                                  	             D   P!          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            *         ,  V         X                                              2         4  ^	         `	  
         
                                   :         <  f         h                                              B         D  n         p                                               J         L  v         x                          !         !  &#         (#  R$         T$  ~%         %  &         &  '         '  )         )  .*         0*  Z+         \+  ,         ,  -         -  .         .  
0         0  61         81  b2         d2  3         3  4         4  5         5  7         7  >8         @8  j9         l9  :         :  ;         ;  <         <  >         >  F?         H?  r@         t@  A         A  B         B  C          `^  ^_          _  `          `  a          a  b          c  d         h  (j         *j  Tk         Vk  l         l  m         m  n         n  p         p  0q         2q  \r         ^r  s         s  t         t  u         u  w         w  8x         :x  dy         fy  z         z  {         {  |         |  ~         ~  @         B  l         n                      Ă                        C         E  o         q             ǉ         ɉ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               0!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-s-     s--    -5-x    5-C-|    C-v-\    v-c0-    c0--h    -Eb-    Eb-x-	    x--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               4!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        ;  Q          v                                    '            <  +          Q  @	          f	  U
          {
  j                                                       >                                 6         7  U         V  t                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |           w                                                          %  	'         
'  9(         :(  i)         j)  *         *  +         +  ,         ,  ).         *.  Y/         Z/  0         0  1         1  2         2  4         4  I5         J5  y6         z6  7         7  8         8  	:         
:  9;         :;  i<         j<  =         =  >         >  ?         ?  )A         *A  YB         ZB  C         C  D         D  E         E  G         G  IH         JH  yI         h  i         i  *k         ,k  Zl         \l  m         m  n         n  o         o  q         q  Jr         Lr  zs         |s  t         t  u         u  
w         w  :x         <x  jy         ly  z         z  {         {  |         |  *~         ,~  Z         \                                              J         L  z         |             ڈ         ܈  
           :         <  j         P  ~                      ސ                      >         @  n         p             Ζ         Ж              .         0  ^         `                                               N         P  ~                      ޣ                      >         @  n         p             Ω         Щ              .         0  ^         `                                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               &!         0!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-s-     s--    -#;-x    #;-I-|    I-}-\    }-=-    =--P    -ib-    ib-c-	    c-g-D
    g-3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               3!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         '  E         F  d         e  	         	  
         
                                                         .          /  	          D            Y  3          n  H            ]                                               I   T!          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            +         ,  W         X                                              3         4  _	         `	  
         
                                   ;         <  g         h                                              C         D  o         p                                               K         L  w         x                          !         !  '#         (#  S$         T$  %         %  &         &  '         '  )         )  /*         0*  [+         \+  ,         ,  -         -  .         .  0         0  71         81  c2         d2  3         3  4         4  5         5  7         7  ?8         @8  k9         l9  :         :  ;         ;  <         <  >         >  G?         H?  s@         t@  A         A  B         B  C          Fd  De          te  rf          f  g          g  h         h  )j         *j  Uk         Vk  l         l  m         m  n         n  p         p  1q         2q  ]r         ^r  s         s  t         t  u         u  w         w  9x         :x  ey         fy  z         z  {         {  |         |  ~         ~  A         B  m         n             Â         Ă                        D         E  p         q             ȉ         ɉ            ʐ  ȑ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               2!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -K -X     K -հ-     հ-ܰ-    ܰ-_/-p    _/-;-T    ;-c-4    c-c0-    c0--8    -Kb-    Kb-x-	    x--8
    -3׳-
                   =         b      J      7   U   7   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               0!         >!  31                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 64 })  outputElements  5184    input0Elements  31104   input1Elements  24576   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     4                           #         b      J      7   M     operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       @                        ;  P          v                                                            +  	         @	  c
         d
                                 P  n         o                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     4                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  1728    input0Elements  1728    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       x                      @  7          `  W            w          %  '         
'  8(         :(  h)         j)  *         *  +         +  ,         ,  (.         *.  X/         Z/  0         0  1         1  2         2  4         4  H5         J5  x6         z6  7         7  8         8  :         
:  8;         :;  h<         j<  =         =  >         >  ?         ?  (A         *A  XB         ZB  C         C  D         D  E         E  G         G  HH         JH  xI         h  i         i  +k         ,k  [l         \l  m         m  n         n  o         o  q         q  Kr         Lr  {s         |s  t         t  u         u  w         w  ;x         <x  ky         ly  z         z  {         {  |         |  +~         ,~  [         \                                              K         L  {         |             ۈ         ܈             ;         <  k         P                        ߐ                      ?         @  o         p             ϖ         Ж              /         0  _         `                                               O         P                        ߣ                      ?         @  o         p             ϩ         Щ              /         0  _         `                                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               (!         4!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,-m-     m--    -;-    ;-_J-    _J-}-l    }-G-    G--P    --    -b-h	    b--p
    -#-
    #--    -9׳-L                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               1!         >!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        =         >  \         ]  {         |            r  ~                                  #  /          ^  j                                                          E  B                                              "         "  1#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        %  1&         2&  ['         \'  (         (  )         )  *         *  ,         ,  --         .-  W.         X.  /         /  0         0  1         1  2          3  )4         *4  S5         T5  }6         ~6  7         7  8         8  9         9  %;         &;  O<         P<  y=         z=  >         >  ?         ?  @         @  !B         "B  KC         LC  uD         vD  E         E  F         F  G         G  I         I  FJ         HJ  pK         rK  L         L  M         M  N         N  P         P  BQ         DQ  lR         nR  S         S  T         T  U         U  W         W  >X         @X  hY         jY  Z         Z  [         [  \         \  ^         ^  :_         <_  d`         f`  a         a  b         b  c         c  e         e  6f         8f  `g         bg  h         h  i          i  j          j  k          
l  	m          4m  3n         ^n  o          o  p         p  q         q  s         s  .t         0t  Xu         Zu  v         v  w         w  x         x   z         z  *{         ,{  T|         V|  ~}         }  ~         ~                        &         (  P         R  z         |             Ά         І             "         $  L         N  v         x             ʍ                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               =!         V!  1         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    f
         g
             9                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     H
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-m-     m--    -#;-x    #;-I-|    I-}-\    }--    --P    -cb-    cb-c-	    c-U-D
    U-9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               1!         <!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 3	         4	  R
         S
  q         ]  {         |                                                         '  !          V  P                                                          9  6           d   a!          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p                        ?         @  _         `                                                        	  
          
  ?         @  _         `                                                                      ?         @  _         `                                                                      ?         @  _         `                          !         !  "         "  #          $  %          %  >&         @&  ^'         `'  ~(         (  )         )  *         *  +         +  ,          -  .          .  >/         @/  ^0         `0  ~1         1  2         2  3         3  4         4  5          6  7          7  >8         @8  ^9         `9  ~:         :  ;         ;  <         <  =         =  >          ?  @          @  >A          F  G         G  
I         I  *J         +J  JK         KK  jL         kL  M         M  N         N  O         O  P         P  
R         R  *S         +S  JT         KT  jU         kU  V         V  W         W  X         X  Y         Y  
[         [  *\         +\  J]         K]  j^         k^  _         _  `         `  a         a  b         b  d         d  6e         7e  Xf         Yf  zg         {g  h          m  n          o  p          Kp  `q          q  r                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               -!         @!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 u
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,-m-     m--    -;-    ;-_J-    _J-}-l    }-G-    G--P    --    -b-h	    b--p
    --
    --    -9׳-L                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               1!         >!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        =         >  \         ]  {         |            r  }                                  #  .          ^  i                                                          E  C                                              "         "  1#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        %  0&         2&  Z'         \'  (         (  )         )  *         *  ,         ,  ,-         .-  V.         X.  /         /  0         0  1         1  2          3  (4         *4  R5         T5  |6         ~6  7         7  8         8  9         9  $;         &;  N<         P<  x=         z=  >         >  ?         ?  @         @   B         "B  JC         LC  tD         vD  E         E  F         F  G         G  I         I  GJ         HJ  qK         rK  L         L  M         M  N         N  P         P  CQ         DQ  mR         nR  S         S  T         T  U         U  W         W  ?X         @X  iY         jY  Z         Z  [         [  \         \  ^         ^  ;_         <_  e`         f`  a         a  b         b  c         c  e         e  7f         8f  ag         bg  h         h  i         i  j         j  	l         
l  3m         4m  ]n         ^n  o         o  p          p  q          q  r          s  t          0t  /u          Zu  Yv         v  w         w  x         x  z         z  +{         ,{  U|         V|  }         }  ~         ~                        '         (  Q         R  {         |             φ         І             #         $  M         N  w         x             ˍ                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               =!         V!  1         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    f
         g
             8                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     I
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-g-     g--    -#;-x    #;-I-|    I-}-\    }-k-    k--P    -cb-    cb-c-	    c-U-D
    U-9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               .!         :!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 3	         4	  R
         S
  q         ]  {         |                                                         '             V  O            ~                                              9  7           d   b!          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p                        >         @  ^         `  ~                                                      	  
          
  >         @  ^         `  ~                                                                    >         @  ^         `  ~                                                                    >         @  ^         `  ~                        !         !  "         "  #          $  %          %  ?&         @&  _'         `'  (         (  )         )  *         *  +         +  ,          -  .          .  ?/         @/  _0         `0  1         1  2         2  3         3  4         4  5          6  7          7  ?8         @8  _9         `9  :         :  ;         ;  <         <  =         =  >          ?  @          @  ?A         G  	I         I  )J         +J  IK         KK  iL         kL  M         M  N         N  O         O  P         P  	R         R  )S         +S  IT         KT  iU         kU  V         V  W         W  X         X  Y         Y  	[         [  )\         +\  I]         K]  i^         k^  _         _  `         `  a         a  b         b  d         d  5e         7e  Wf         Yf  yg         {g  h          r  s          t  )u          Wu  lv          v  w          w  x                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               -!         @!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 u
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,-y-     y--    -5-    5-;D-    ;D-w-l    w-12-    12--x    -ob-	    ob--
    -#-
    #-y-|    y-ճ-    ճ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               3!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         +  (          V  S            ~                                              5  6	          d	  e
          
                                               =         >  \         H  m         n                                                                 b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         D  S          ~                                    ,  ;         $  %         K  L         L  N         N  EO         FO  mP         nP  Q         Q  R         R  S         S  U         U  5V         6V  ]W         ^W  X         X  Y         Y  Z         Z  [         [  %]         &]  M^         N^  u_         v_  `         `  a         a  b         b  d         d  =e         >e  ef         ff  g         g  h         h  i         i  k         k  -l         .l  Um         Vm  }n         ~n  o         o  p         p  q         q  s         s  Et         Ft  mu         nu  v         v  w         w  x         x  z         z  5{         6{  ]|         ^|  }         }  ~         ~                        %         &  M         N  u         v             ņ         Ɔ                        =         >  e         f                        ݎ         ގ             -         .  T         V  |         ~             ̕         Ε                        D         F  l         n                                              4         6  \         ^                        Ԥ         ֤             $         &  L         N  t         v             ī         ƫ                        <         >  d         f                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               4!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    f
         g
             9      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    f
         g
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     d
            4     -,-`     ,--     --    -7<-    7<-J-    J-~-d    ~--    --p    -b-    b-!-
    !-/-
    /-۰-t    ۰-9׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               :!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        =         >  \         ]  {         |                                                C  @          n  k                                               J  [                                               C!         D!  q"         r"  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                     <  E          p  y                                 $  8%         :%  [&         \&  }'         ~'  (         (  )         )  *         *  ,         ,  '-         (-  I.         J.  k/         l/  0         0  1         1  2         2  3         3  5         5  76         86  Y7         Z7  {8         |8  9         9  :         :  ;         ;  =         =  %>         &>  G?         H?  i@         j@  A         A  B         B  C         C  D         D  F         F  5G         6G  VH         XH  xI         zI  J         J  K         K  L         L  "N         $N  UO         WO  P         P  Q         Q  R         R  !T         #T  TU         VU  V         V  W         W  X         X  'Z         (Z  ][         ^[  \         \  ]         ]  ^          _  5`         6`  ka         la  b         b  c         c  e         e  ?f         @f  sg         tg  h         h  i         i  k         k  Cl         Dl  wm         xm  n         n  o         o   q         q  "r         $r  Ds         Fs  ft         ht  u         u  v         v  w         w  x         x  z         z  2{         4{  T|         V|  v}         x}  ~         ~             ܀         ހ                        "  B         D  d         f                        ʈ         ̈                        0                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               :!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    f
         g
             C      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     3
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,-y-     y--    -5-    5-5D-    5D-w-l    w-12-    12--x    -ib-	    ib--
    -#-
    #-y-|    y-ճ-    ճ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               3!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         +  )          V  T                                                          5  7	          d	  f
          
                                               =         >  \         H  l         m                                                                  b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         f  u                                    #          N  ]         $  %         K  L         L  N         N  DO         FO  lP         nP  Q         Q  R         R  S         S  U         U  4V         6V  \W         ^W  X         X  Y         Y  Z         Z  [         [  $]         &]  L^         N^  t_         v_  `         `  a         a  b         b  d         d  <e         >e  df         ff  g         g  h         h  i         i  k         k  ,l         .l  Tm         Vm  |n         ~n  o         o  p         p  q         q  s         s  Dt         Ft  lu         nu  v         v  w         w  x         x  z         z  4{         6{  \|         ^|  }         }  ~         ~                        $         &  L         N  t         v             Ć         Ɔ                        <         >  d         f                        ܎         ގ             ,         .  U         V  }         ~             ͕         Ε                        E         F  m         n                                              5         6  ]         ^                        դ         ֤             %         &  M         N  u         v             ū         ƫ                        =         >  e         f                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               2!         @!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    f
         g
             8      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    f
         g
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     c
            4     -,-`     ,--     -#-    #-7<-    7<-J-    J-~-d    ~--    --p    -{b-    {b-!-
    !-/-
    /-۰-t    ۰-9׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               <!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        =         >  \         ]  {         |                                                C  A          n  l                                              J  Z                                               A!         B!  p"         q"  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                     @  I          t  }                                 $  9%         :%  Z&         \&  |'         ~'  (         (  )         )  *         *  ,         ,  &-         (-  H.         J.  j/         l/  0         0  1         1  2         2  3         3  5         5  66         86  X7         Z7  z8         |8  9         9  :         :  ;         ;  =         =  $>         &>  F?         H?  h@         j@  A         A  B         B  C         C  D         D  F         F  4G         6G  WH         XH  yI         zI  J         J  K         K  L         L  #N         $N  VO         WO  P         P  Q         Q  R         R  "T         #T  UU         VU  V         V  W         W  X         X  &Z         (Z  \[         ^[  \         \  ]         ]  ^          _  4`         6`  ja         la  b         b  c         c  
e         e  >f         @f  rg         tg  h         h  i         i  k         k  Bl         Dl  vm         xm  n         n  o         o  q         q  #r         $r  Es         Fs  gt         ht  u         u  v         v  w         w  x         x  z         z  3{         4{  U|         V|  w}         x}  ~         ~             ݀         ހ              !         "  C         D  e         f                        ˈ         ̈                        1                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               8!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    f
         g
             D      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     4
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -;-    ;-;-x    ;-kJ-|    kJ-}-\    }-4-    4--h    -b-    b-	n-	    	n--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               G!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        =         >  \         ]  {         |            r  s                                  #  $          ^  _                                                          E  B                                              "         "  1#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            $         &  J         L  p         r                                            
  .	         0	  T
         V
  z         |                                              8         :  ^         `                                                         B         D  h         j                                               &          (   L!         N!  r"          ?  +@          Y@  nA          A  B          B  C          "D  7E         K  L         L  M          N  %O         &O  KP         LP  qQ         rQ  R         R  S         S  T         T  	V         
V  /W         0W  UX         VX  {Y         |Y  Z         Z  [         [  \         \  ^         ^  9_         :_  _`         ``  a         a  b         b  c         c  d         d  f         f  Cg         Dg  ih         jh  i         i  j         j  k         k  m         m  'n         (n  Mo           ?         @  e         f                        י         ؙ             #         $  I         J  o         p                                              -         .  S         T  y         z             Ũ         ƨ                        7         8  ]         ^                        ϰ         а                        A         B  g                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               H!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     %
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    <     -,-p     ,--     -5-    5-)5-    )5-C-    C-w-t    w-5-     5-5=-X    5=--    -b-p	    b--x
    -)-      )--     -ճ-     ճ--z                    =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               E!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        ;  I          v                                    '            <            Q  0	          f	  E
          {
  Z                                                       *                      +         ,  J         K  i         j                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        8"  Y#         [#  |$         ~$  %         %  &         &  '         '  )         )  0*         2*  T+         V+  x,         z,  -         -  .         .  /         /  1         
1  ,2         .2  P3         R3  t4         v4  5         5  6         6  7         7  9         9  (:         *:  L;         N;  p<         r<  =         =  >         >  ?         ?   A         A  $B         &B  HC         JC  lD         nD  E         E  F         F  G         -l  Nm         Pm  qn         sn  o         o  p         p  q         q  s         s  %t         't  Iu         Ku  mv         ov  w         w  x         x  y         y  z         z  !|         #|  E}         G}  i~         k~                                    ׁ  Ђ                                 C  e         g                        щ         Ӊ           ړ                       "  D         F  h         j                        ԛ         ֛                        @         B  d         f                        У         ң                        <         >  `         b                        ̫         Ϋ                        8         :  \         ^                        ȳ         ʳ                                           b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               D!         H!  1         1  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   b      operation.2     exchange                 :               *      :      R      reason     z   OperatorInputs     2   value         6          b   invocations           1          7                f
         g
             4                                    b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                          b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    f
         g
                                      b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                                 
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     S
            0     -,-X     ,--     -;-    ;-;-x    ;-kJ-|    kJ-}-\    }-4-    4--h    -b-    b-	n-	    	n--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               G!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        =         >  \         ]  {         |            r  r                                  #  #          ^  ^                                                          E  C                                              "         "  1#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            %         &  K         L  q         r                                   	         
  /	         0	  U
         V
  {         |                                              9         :  _         `                                                         C         D  i         j                                              '          (   M!         N!  s"          eE  zF          F  G          G   I          .I  CJ          qJ  K         K  L         L  M          N  $O         &O  JP         LP  pQ         rQ  R         R  S         S  T         T  V         
V  .W         0W  TX         VX  zY         |Y  Z         Z  [         [  \         \  ^         ^  8_         :_  ^`         ``  a         a  b         b  c         c  d         d  f         f  Bg         Dg  hh         jh  i         i  j         j  k         k   m         m  &n         (n  Lo           >         @  d         f                        ֙         ؙ             "         $  H         J  n         p                                              ,         .  R         T  x         z             Ĩ         ƨ                        6         8  \         ^                        ΰ         а                        @         B  f                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               F!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     $
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    <     -,-p     ,--     -;-    ;-/5-    /5-C-    C-w-t    w-5-     5-5=-X    5=--    -b-p	    b--x
    -)-
    )--    -ճ-\    ճ--<                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               I!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        ;  H          v                                    '            <            Q  /	          f	  D
          {
  Y                                                       *                      0         1  O         P  n         o                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        8"  Z#         [#  }$         ~$  %         %  &         &  '         '  )         )  1*         2*  U+         V+  y,         z,  -         -  .         .  /         /  	1         
1  -2         .2  Q3         R3  u4         v4  5         5  6         6  7         7  9         9  ):         *:  M;         N;  q<         r<  =         =  >         >  ?         ?  A         A  %B         &B  IC         JC  mD         nD  E         E  F         F  G         -l  Om         Pm  rn         sn  o         o  p         p  q         q  s         s  &t         't  Ju         Ku  nv         ov  w         w  x         x  y         y  z         z  "|         #|  F}         G}  j~         k~                        ց         ׁ                        B          C  <          g  `                                  Ӊ  ̊         ړ             !         "  E         F  i         j                        ՛         ֛                        A         B  e         f                        ѣ         ң                        =         >  a         b                        ͫ         Ϋ                        9         :  ]         ^                        ɳ         ʳ                                           b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               F!         H!  1         1  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    f
         g
             4                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    f
         g
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     T
            0     -,-X     ,--     -5-    5-5-x    5-D-|    D-kw-\    kw--    --h    -b-    b-n-	    n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               G!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         +  (          V  S            ~                                              5  1	          d	  `
          
                                               =         >  \         H  f         g                                                                 b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            "         $  F         H  j         l                                              	          	  B
         D
  f         h                                                         >         @  b         d                                                         :         <  ^         `                                              !         !  6"          t"  l#          #  $          $  %          %  &          '  (         .I  PJ         RJ  tK         vK  L         L  M         M  N         N  P         P  (Q         *Q  LR         NR  pS         rS  T         T  U         U  V         V   X         X  $Y         &Y  HZ         JZ  l[         n[  \         \  ]         ]  ^         ^  _         _  #a         $a  Ib         Jb  oc         pc  d         d  e         e  f         f  h         h  -i         .i  Sj         Tj  yk         m  n         n  p         p  <q         =q  `r         ar  s         s  t         t  u         u  v         v  x         x  8y         9y  \z         ]z  {         {  |         |  }         }  ~         ~             4         5  X         Y  |         }             ą         Ņ                        0         1  T         U  x         y                                                      >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               G!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     8
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-;-x    ;-kJ-|    kJ-}-\    }--    --h    -b-    b-3n-	    3n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               @!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        =         >  \         ]  {         |                                                C  @          n  k                                              J  X                                                          "         "  1#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            $         &  J         L  p         r                                            
  .	         0	  T
         V
  z         |                                              8         :  ^         `                                                         B         D  h         j                                               &          (   L!         N!  r"         t"  #         #  $         $  %         %  '         '  1(         2(  W)         X)  }*         ~*  +         +  ,         ,  -         -  /         /  ;0         <0  a1         b1  2          2  3          3  4          4  5          $6  *7          X7  ^8         _j  k         k  l         l  m         m  n         n  p         p  Aq         Cq  gr         ir  s         s  t         t  u         u  v         w  %x         'x  Ky         My  qz         sz  {         {  |         |  }         }  	           /         1  U         W  {         }             Ǆ         Ʉ                        9         ;  _         a                        ь         ӌ                      	  .         /  T         U  z         {             ƕ         Ǖ                        8         9  ^         _                        Н         ѝ                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               A!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     O
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-5-x    5-D-|    D-kw-\    kw--    --h    -b-    b-n-	    n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               G!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         +  )          V  T                                                          5  0	          d	  _
          
                                               =         >  \         H  f         g                                                                 b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            #         $  G         H  k         l                                              	          	  C
         D
  g         h                                                         ?         @  c         d                                                         ;         <  _         `                                              !         !  7"          2(  *)          X)  P*          ~*  v+          +  ,          ,  -         .I  QJ         RJ  uK         vK  L         L  M         M  N         N  P         P  )Q         *Q  MR         NR  qS         rS  T         T  U         U  V         V  X         X  %Y         &Y  IZ         JZ  m[         n[  \         \  ]         ]  ^         ^  _         _  "a         $a  Hb         Jb  nc         pc  d         d  e         e  f         f  h         h  ,i         .i  Rj         Tj  xk         m  n         n  p         p  ;q         =q  _r         ar  s         s  t         t  u         u  v         v  x         x  7y         9y  [z         ]z  {         {  |         |  }         }  ~         ~             3         5  W         Y  {         }             Å         Ņ                        /         1  S         U  w         y                                                      >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               C!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     7
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-;-x    ;-kJ-|    kJ-}-\    }--    --h    -b-    b--n-	    -n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               B!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        =         >  \         ]  {         |                                                C  A          n  l                                              J  W                                                          "         "  1#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            %         &  K         L  q         r                                   	         
  /	         0	  U
         V
  {         |                                              9         :  _         `                                                         C         D  i         j                                              '          (   M!         N!  s"         t"  #         #  $         $  %         %  
'         '  0(         2(  V)         X)  |*         ~*  +         +  ,         ,  -         -  /         /  :0         <0  `1         b1  2          8  9          9  :          :  ;          (<  .=          \=  b>         _j  k         k  l         l  m         m  n         n  p         p  Bq         Cq  hr         ir  s         s  t         t  u         u   w         w  &x         'x  Ly         My  rz         sz  {         {  |         |  }         }  
           0         1  V         W  |         }             Ȅ         Ʉ                        :         ;  `         a                        Ҍ         ӌ                      	  -         /  S         U  y         {             ŕ         Ǖ                        7         9  ]         _                        ϝ         ѝ                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               E!         R!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     N
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -;-    ;-;-x    ;-I-|    I-|-\    |-+-    +--h    -ib-    ib-x-	    x--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               I!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         '  E         F  d         e  	         	  
         
           g                                                                    !            6  !          K  6          `  K          u                                    &   :!          a!  u"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                      "  B         D  d         f                                              	         	  0
         2
  R         T  t         v                                                          @         B  b         d                                                         .         0  P         R  r         t                                     !         "  1#         3#  S$          q0  |1          1  2          2  3          4  5          I5  T6         sG  H         H  I         I  J         J  K         K  M         M  =N         ?N  _O         aO  P         P  Q         Q  R         R  S         S  	U         U  +V         -V  MW         OW  oX         qX  Y         Y  Z         Z  [         [  \         \  ^         ^  ;_         =_  ]`         _`  a         a  b         b  c         c  d         d  f         	f  )g         +g  Kh         Mh  mi         oi  j         j  k         k  l         $  D         F  f         h                        ̕         Ε                        2         4  T         V  v         x                        ܞ         ޞ                        "  B         D  d         f                        ʦ         ̦                        0         2  R         T  t                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               2!         >!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 q
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-#5-x    #5-C-|    C-w-\    w-I-    I--h    -ub-    ub--n-	    -n--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               <!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        ;  I          v                                    '            <            Q  0	          f	  E
          {
  Z                                                       *                      +         ,  J         K  i         j                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            &         (  N         P  v         x                                              >	         @	  f
         h
                                              .         0  V         X  ~                                                        F         H  n         p                                              6         8  ^          `   !         !  "         "  $         $  :%         <%  b&         d&  '         '  (         (  )         )  +         +  *,         ,,  R-         T-  z.         |.  /         /  0         0  1         1  3         3  B4         D4  j5         l5  6         6  7         7  8         8  
:         :  2;         4;  Z<         \<  =         =  >         >  ?         ?  @         @  "B         $B  JC         LC  rD         tD  E         E  F          S  T          T  V          =V  OW          W  X          X  Y         Bm  hn         jn  o         o  p         p  q         q  s         
s  0t         2t  Xu         Zu  v         v  w         w  x         x  y         y   {         "{  H|         J|  p}         r}  ~         ~                                   8         :  `         b                        ؇         ڇ              (         *  P         R  x         z             Ȏ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               8!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     O
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-;-x    ;-I-|    I-|-\    |-+-    +--h    -ob-    ob-x-	    x--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               E!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         '  E         F  d         e  	         	  
         
           l                                                                     +            @  *          U  ?          j  T                                               0   C!          k!  ~"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            !         "  C         D  e         f                                              	         	  1
         2
  S         T  u         v                                                          A         B  c         d                                                         /         0  Q         R  s         t                                     !         "  2#         3#  T$          6  7          7  8          8  9          !:  ,;          W;  b<         sG  H         H  I         I  J         J  K         K  M         M  >N         ?N  `O         aO  P         P  Q         Q  R         R  S         S  
U         U  ,V         -V  NW         OW  pX         qX  Y         Y  Z         Z  [         [  \         \  ^         ^  <_         =_  ^`         _`  a         a  b         b  c         c  d         d  f         	f  *g         +g  Lh         Mh  ni         oi  j         j  k         k  l         $  E         F  g         h                        ͕         Ε                        3         4  U         V  w         x                        ݞ         ޞ              !         "  C         D  e         f                        ˦         ̦                        1         2  S         T  u                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               4!         B!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 q
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-)5-x    )5-C-|    C-w-\    w-I-    I--h    -ub-    ub--n-	    -n--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               >!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        ;  H          v                                    '            <            Q  /	          f	  D
          {
  Y                                                       *                      0         1  O         P  n         o                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            '         (  O         P  w         x                                              ?	         @	  g
         h
                                              /         0  W         X                                                          G         H  o         p                                              7         8  _          `   !         !  "         "  $         $  ;%         <%  c&         d&  '         '  (         (  )         )  +         +  +,         ,,  S-         T-  {.         |.  /         /  0         0  1         1  3         3  C4         D4  k5         l5  6         6  7         7  8         8  :         :  3;         4;  [<         \<  =         =  >         >  ?         ?  @         @  #B         $B  KC         LC  sD         tD  E         E  F          Z  [          I[  [\          \  ]          ]  ^          _  $`         Bm  in         jn  o         o  p         p  q         q  	s         
s  1t         2t  Yu         Zu  v         v  w         w  x         x  y         y  !{         "{  I|         J|  q}         r}  ~         ~                                   9         :  a         b                        ه         ڇ             )         *  Q         R  y         z             Ɏ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               :!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     N
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     --    -w5-x    w5-D-|    D-_w-\    _w--    --h    -Kb-    Kb-x-	    x--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ;!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         +  (          V  S            ~                                              5  1	          d	  `
          
                                               =         >  \         H  f         g                                                                 b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                              P  L            |         "  $         $  *%         ,%  J&         L&  j'         l'  (         (  )         )  *         *  +         +  
-         -  *.         ,.  J/         L/  j0         l0  1         1  2         2  3         3  4         4  
6         6  *7         ,7  J8         L8  j9         l9  :         :  ;         ;  <         <  =         =  
?         ?  *@         ,@  JA         LA  jB         lB  C         C  D         D  E         AF  _G         aG  H         H  I         I  J         J  K         K  L         M  N         !N  ?O         AO  _P         aP  Q         Q  R         R  S         S  T         T  U         V  W         !W  ?X         AX  _Y         aY  Z         Z  [         [  \         \  ]         ]  ^         _  `         !`  ?a         Aa  _b         ab  c         c  d         d  e         e  f         f  g         h  i         i  k         k  1l         2l  Qm         Rm  qn         rn  o         o  p         p  q         q  r         r  t         t  1u         2u  Qv         Rv  qw         rw  x         x  y         y  z         z  {         {  }         }  1~         2~  Q         R  q         r                        у         ҃                        1         2  Q         R  q                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               )!         0!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 q
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-#;-x    #;-I-|    I-}-\    }-o-    o--h    -ob-    ob-x-	    x--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               :!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 3	         4	  R
         S
  q         ]  {         |                                                         '  )          V  X                                                          9  6           d   a!          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            )         *  S         T  }         ~                                   %         &  O	         P	  y
         z
                                   !         "  K         L  u         v                                              G         H  q         r                                              C         D  m         n               !         !  "         $  <%         >%  f&         h&  '         '  (         (  )         )  +         +  8,         :,  b-         d-  .         .  /         /  0         0  
2         2  43         63  ^4         `4  5         5  6         6  7         7  9         9  0:         2:  Z;         \;  <         <  =         =  >         >  @         @  ,A         .A  VB         XB  C         C  D         D  E         E  F          G  (H         :H  bI         dI  J         J  K         K  L         L  
N         N  4O         6O  ^P         `P  Q         Q  R         R  S         S  U         U  0V         2V  ZW         \W  X         X  Y         Y  Z         Z  \         \  ,]         .]  V^         X^  _         _  `         `  a         a  b          c  (d         *d  Re         Te  |f         ~f  g         g  h          :  4          b  \                                  ڇ  Ԉ         ʎ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               3!         F!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -W -X     W -հ-     հ-ܰ-    ܰ-/-p    /-q;-T    q;-c-4    c--    --8    -Qb-    Qb-x-	    x--8
    -9׳-
                   =         b      J      7   U   7   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               7!         >!  :1                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 64 })  outputElements  5184    input0Elements  31104   input1Elements  24576   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     4                           #         b      J      7   M     operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       @                         +  )          V  T                                              
            9  4	         h	  
         
                                                                                                  b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     4                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  1728    input0Elements  1728    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       x                                    !          @!  <"         "  
$         $  +%         ,%  K&         L&  k'         l'  (         (  )         )  *         *  +         +  -         -  +.         ,.  K/         L/  k0         l0  1         1  2         2  3         3  4         4  6         6  +7         ,7  K8         L8  k9         l9  :         :  ;         ;  <         <  =         =  ?         ?  +@         ,@  KA         LA  kB         lB  C         C  D         D  E         AF  `G         aG  H         H  I         I  J         J  K         K   M         M   N         !N  @O         AO  `P         aP  Q         Q  R         R  S         S  T         T   V         V   W         !W  @X         AX  `Y         aY  Z         Z  [         [  \         \  ]         ]   _         _   `         !`  @a         Aa  `b         ab  c         c  d         d  e         e  f         f   h         h  i         i  k         k  0l         2l  Pm         Rm  pn         rn  o         o  p         p  q         q  r         r  t         t  0u         2u  Pv         Rv  pw         rw  x         x  y         y  z         z  {         {  }         }  0~         2~  P         R  p         r                        Ѓ         ҃                        0         2  P         R  p                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               +!         4!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 q
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     --    -#;-x    #;-I-|    I-}-\    }-o-    o--h    -ub-    ub-x-	    x--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               8!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 3	         4	  R
         S
  q         ]  {         |                                                         '  (          V  W                                                          9  7           d   b!          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            (         *  R         T  |         ~                                   $         &  N	         P	  x
         z
                                             "  J         L  t         v                                              F         H  p         r                                              B         D  l         n               !         !  "         $  =%         >%  g&         h&  '         '  (         (  )         )  +         +  9,         :,  c-         d-  .         .  /         /  0         0  2         2  53         63  _4         `4  5         5  6         6  7         7  9         9  1:         2:  [;         \;  <         <  =         =  >         >  @         @  -A         .A  WB         XB  C         C  D         D  E         E  F          G  )H         :H  cI         dI  J         J  K         K  L         L  N         N  5O         6O  _P         `P  Q         Q  R         R  S         S  U         U  1V         2V  [W         \W  X         X  Y         Y  Z         Z  \         \  -]         .]  W^         X^  _         _  `         `  a         a  b          c  )d         *d  Se         Te  }f         ~f  g         g  h                      *  $          R  L          z  t                     ʎ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               7!         J!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-m-     m--    -;-x    ;-I-|    I-|-\    |-
-    
--P    -Wb-    Wb-c-	    c-y-D
    y-3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               2!         <!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         '  E         F  d         e  	         	  
         
           g                                                                    !            6  !          K  6          `  K          u                                    &   :!          a!  u"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            .         0  ^         `                                               N         P  ~	         	  
         
                        >         @  n         p                                    .         0  ^         `                                               N         P  ~                                              >!         @!  n"          $  %         %  &         &  ((         *(  X)         Z)  *         *  +         +  ,         ,  .         .  H/         J/  x0         z0  1         1  2         2  4         
4  85         :5  h6         j6  7         7  8         8  9         9  (;         *;  X<         Z<  =         =  >         >  ?         ?  A         A  HB         JB  xC         zC  D         D  E         E  G         
G  8H          :H  <I          dI  fJ          J  K          K  L         h   j         j  0k         2k  `l         bl  m         m  n         n  o         o   q         "q  Pr         Rr  s         s  t         t  u         u  w         w  @x         Bx  py         ry  z         z  {         {   }         }  0~         2~  `         b                                              L         N  {         }             ۈ         ݈             ;                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               *!         @!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     --    -5-x    5-C-|    C-w-\    w--    --h    -Qb-    Qb-x-	    x--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               :!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        ;  I          v                                    '            <            Q  0	          f	  E
          {
  Z                                                       *                      +         ,  J         K  i         j                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                        "            D  /          f  Q            s         &  '         '  (         (  
*         *  6+         7+  b,         c,  -         -  .         .  /         /  1         1  >2         ?2  j3         k3  4         4  5         5  6         6  8         8  F9         G9  r:         s:  ;         ;  <         <  =         =  !?         #?  M@         O@  yA         {A  B         B  C         C  D         D  )F         +F  UG         WG  H         H  I         I  J         J  L         L  1M         3M  ]N         _N  O         O  P         P  Q         Q  S         S  9T         ;T  eU         gU  V         V  W         W  X         X  Z         Z  A[         C[  m\         o\  ]         ]  ^         ^  _         _  a         a  Ib         Kb  uc         wc  d         d  e         e  f         f  %h         'h  Qi         Si  }j         j  k         k  l          n  *o         ,o  Vp         Xp  q         q  r         r  s         s  u         u  2v         4v  ^w         `w  x         x  y         y  z         z  |         |  :}         <}  f~         h~                                              B         D  n         p             Ƈ         ȇ                         J         L  v         x             Ύ         Ў             &                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               *!         8!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-g-     g--    -;-x    ;-I-|    I-|-\    |-
-    
--P    -Wb-    Wb-c-	    c-y-D
    y-3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               /!         :!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         '  E         F  d         e  	         	  
         
           l                                                                     +            @  *          U  ?          j  T                                               0   C!          k!  ~"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            /         0  _         `                                               O         P  	         	  
         
                        ?         @  o         p                                    /         0  _         `                                               O         P                                                ?!         @!  o"         %  &         &  )(         *(  Y)         Z)  *         *  +         +  ,         ,  .         .  I/         J/  y0         z0  1         1  2         2  	4         
4  95         :5  i6         j6  7         7  8         8  9         9  );         *;  Y<         Z<  =         =  >         >  ?         ?  A         A  IB         JB  yC         zC  D         D  E         E  	G         
G  9H          L  M          N  O          6O  8P          `P  bQ          Q  R         h  j         j  1k         2k  al         bl  m         m  n         n  o         o  !q         "q  Qr         Rr  s         s  t         t  u         u  w         w  Ax         Bx  qy         ry  z         z  {         {  }         }  1~         2~  a         b                                              M         N  |         }             ܈         ݈             <                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               *!         @!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-y-     y--    -5-x    5-C-|    C-w-\    w--    --h    -Wb-    Wb-x-	    x--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               8!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        ;  H          v                                    '            <            Q  /	          f	  D
          {
  Y                                                       *                      0         1  O         P  n         o                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                             	  	          2
           &  '         '  (         (  	*         *  5+         7+  a,         c,  -         -  .         .  /         /  1         1  =2         ?2  i3         k3  4         4  5         5  6         6  8         8  E9         G9  q:         s:  ;         ;  <         <  =         =  "?         #?  N@         O@  zA         {A  B         B  C         C  D         D  *F         +F  VG         WG  H         H  I         I  J         J  L         L  2M         3M  ^N         _N  O         O  P         P  Q         Q  S         S  :T         ;T  fU         gU  V         V  W         W  X         X  Z         Z  B[         C[  n\         o\  ]         ]  ^         ^  _         _  a         a  Jb         Kb  vc         wc  d         d  e         e  f         f  &h         'h  Ri         Si  ~j         j  k         k  l          n  +o         ,o  Wp         Xp  q         q  r         r  s         s  u         u  3v         4v  _w         `w  x         x  y         y  z         z  |         |  ;}         <}  g~         h~                                              C         D  o         p             Ǉ         ȇ                         K         L  w         x             ώ         Ў             '                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               .!         <!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,-s-     s--    -;-    ;-eJ-    eJ-}-d    }--    --p    -ob-    ob-	-
    	--
    --t    --׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               9!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                       !  A         B  b         c                       |                                    -  A          h  |                                              $  !          O  L                                              "         "  ;#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         "            ?  1          \  N          y  k                     %  &         &  '         '  )         
)  6*         8*  d+         f+  ,         ,  -         -  .         .  0         0  J1         L1  x2         z2  3         3  4         4  6         6  07         27  ^8         `8  9         9  :         :  ;         ;  =         =  D>         F>  r?         t?  @         @  A         A  B         B  *D         ,D  XE         ZE  F         F  G         G  H         H  J         J  >K         @K  lL         nL  M         M  N         N  O         O  %Q         &Q  SR         TR  S         S  T         T  U         U  W         W  9X         :X  gY         hY  Z         Z  [         [  \         \  ^          ^  M_         N_  {`         |`  a         a  b         b  d         d  3e         4e  af         bf  g         g  h         h  i         i  k         k  Gl         Hl  um         vm  n         n  o         o  p          q  ,r         .r  Zs         \s  t         t  u         u  v         v  x         x  @y         By  nz         pz  {         {  |         |  }         }  &         (  T         V                        ރ                      :         <  h         j             ĉ         Ɖ                       "  N         P  |         ~                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               7!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    f
         g
             9      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     P
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-O-     O--    -;-x    ;-I-|    I-}-\    }-,-    ,--h    -cb-    cb--	    --h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               #!         4!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 3	         4	  R
         S
  q         ]  {         |                                                         '  )          V  X                                                          9  6           d   a!          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            -         .  [         \                                              A         B  o	         p	  
         
                        '         (  U         V                                              ;         <  i         j                                   !         "  O         P  }         ~                        !         !  5"         6"  c#         %  &         &  '         '  %)         &)  S*         T*  +         +  ,         ,  -         -  /         /  90         :0  g1         h1  2         2  3         3  4         4  6          6  M7         N7  {8         |8  9         9  :         :  <         <  3=         4=  a>         b>  ?         ?  @         @  A         A  C         C  GD         HD  uE         vE  F         F  G         G  H         TI  J         J  K          0X  BY          pY  Z          Z  [          [  ]          0]  B^           ?         A  m         o             ɐ         ː             %         '  S         U                        ݗ         ߗ             9         ;  g         i             Ý         ŝ                      !  M         O  {         }             פ         ٤             3         5  a         c                                                      >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               1!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   +      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,-y-     y--    -;-    ;-kJ-    kJ-}-d    }--    --p    -ob-    ob-	-
    	--
    --t    --׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ;!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             !         "  C         D  e         f                                                 
          2  E          m                                                )  '          T  R                                 !         !  !"         ""  @#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                             
            '           %  &         &  '         '  	)         
)  7*         8*  e+         f+  ,         ,  -         -  .         .  0         0  K1         L1  y2         z2  3         3  4         4  6         6  17         27  _8         `8  9         9  :         :  ;         ;  =         =  E>         F>  s?         t?  @         @  A         A  B         B  +D         ,D  YE         ZE  F         F  G         G  H         H  J         J  ?K         @K  mL         nL  M         M  N         N  O         O  $Q         &Q  RR         TR  S         S  T         T  U         U  
W         W  8X         :X  fY         hY  Z         Z  [         [  \         \  ^          ^  L_         N_  z`         |`  a         a  b         b  d         d  2e         4e  `f         bf  g         g  h         h  i         i  k         k  Fl         Hl  tm         vm  n         n  o         o  p          q  -r         .r  [s         \s  t         t  u         u  v         v  x         x  Ay         By  oz         pz  {         {  |         |  }         }  '         (  U         V                        ߃                      ;         <  i         j             ŉ         Ɖ             !         "  O         P  }         ~                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               7!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    f
         g
             8      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     Q
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-O-     O--    -;-x    ;-I-|    I-}-\    }-,-    ,--h    -Wb-    Wb--	    --h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               "!         2!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 3	         4	  R
         S
  q         ]  {         |                                                         '  (          V  W                                                          9  7           d   b!          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ,         .  Z         \                                              @         B  n	         p	  
         
                        &         (  T         V                                              :         <  h         j                                             "  N         P  |         ~                        !         !  4"         6"  b#         %  &         &  '         '  $)         &)  R*         T*  +         +  ,         ,  -         -  
/         /  80         :0  f1         h1  2         2  3         3  4         4  6          6  L7         N7  z8         |8  9         9  :         :  <         <  2=         4=  `>         b>  ?         ?  @         @  A         A  C         C  FD         HD  tE         vE  F         F  G         G  H         TI  J         J  K          p^  _          _  `          `  b          0b  Bc          pc  d           @         A  n         o             ʐ         ː             &         '  T         U                        ޗ         ߗ             :         ;  h         i             ĝ         ŝ                       !  N         O  |         }             ؤ         ٤             4         5  b         c                                                      >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               -!         @!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   ,      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    <     -,-p     ,-s-     s--    -k5-    k5-D-    D-Sw-t    Sw-4-     4-;-X    ;--    -ib-p	    ib-	-x
    	--
    -g-    g-ճ-\    ճ--<                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               7!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         +  (          V  S            ~                                              5  1	          d	  `
          
                                               =         >  \         H  f         g                                                                 b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        D%  n&         JL  uM         vM  N         N  O         O  P         P  %R         &R  QS         RS  }T         ~T  U         U  V         V  X         X  -Y         .Y  YZ         ZZ  [         [  \         \  ]         ]  	_         
_  5`         6`  aa         ba  b         b  c         c  d         d  f         f  =g         >g  ih         jh  i         i  j         j  k         k  m         m  En         Fn  qo         ro  p         p  q         q  r         r  !t         "t  Mu         Nu  yv         zv  w         w  x         x  y         y  ){         *{  U|         V|  }         }  ~         ~                        1         2  ]         ^                                              9         :  e         f                                              A         B  m         n             ē         Ɠ                        H         J  t         v             ̚         Κ             $         &  P         R  |         ~             ԡ         ֡              ,         .  X         Z                        ܨ          ި  ߩ          
            6  7          b  c                                             <         >  h         j                                b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               8!         H!  1         1  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    f
         g
             9                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    f
         g
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            0     -,-X     ,-y-     y--    -;-x    ;-SJ-|    SJ-}-\    }-u-    u--h    -ob-    ob--	    --h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               :!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        =         >  \         ]  {         |                                                C  @          n  k                                              J  X                                                          "         "  1#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            .         0  ^         `                                               N         P  ~	         	  
         
                        >         @  n         p                                    .         0  ^         `                                               N         P  ~                                              >!         @!  n"         p"  #         D%  s&         t&  '         '  (         (  *         *  3+         4+  c,         d,  -         -  .         .  /         /  #1         $1  S2         T2  3         3  4         4  5         5  7         7  C8         D8  s9         t9  :         :  ;         ;  =         =  3>         4>  c?         d?  @         @  A         A  B         B  #D         $D  SE         TE  F         F  G         G  H          e  f          f  g          h  !i          Oi  dj          j  k         ir  s         s  t         t  u         u  'w         )w  Wx         Yx  y         y  z         z  {         {  }         }  G~         I~  w         y             ׁ         ف           	  7         9  g         i             Ǉ         ɇ             '         )  W         Y                                              G         I  w         y             ה         ٔ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               @!         J!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   +      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    <     -,-p     ,-y-     y--    -k5-    k5-D-    D-Sw-t    Sw-3-     3-y;-X    y;--    -ib-p	    ib--x
    --
    -g-    g-ճ-\    ճ--<                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               9!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         +  )          V  T                                                          5  0	          d	  _
          
                                               =         >  \         H  f         g                                                                 b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        D%  o&         JL  tM         vM  N         N  O         O  P         P  $R         &R  PS         RS  |T         ~T  U         U  V         V   X         X  ,Y         .Y  XZ         ZZ  [         [  \         \  ]         ]  _         
_  4`         6`  `a         ba  b         b  c         c  d         d  f         f  <g         >g  hh         jh  i         i  j         j  k         k  m         m  Dn         Fn  po         ro  p         p  q         q  r         r   t         "t  Lu         Nu  xv         zv  w         w  x         x  y         y  ({         *{  T|         V|  }         }  ~         ~                        0         2  \         ^                                              8         :  d         f                                              @         B  l         n             œ         Ɠ                        I         J  u         v             ͚         Κ             %         &  Q         R  }         ~             ա         ֡             -         .  Y         Z                        ݨ         ި  	         
  5         6  a         b                                                           >  ?          j  k                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               :!         H!  1         1  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    f
         g
             8                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    f
         g
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            0     -,-X     ,--     --    -;-x    ;-SJ-|    SJ-}-\    }-u-    u--h    -ob-    ob--	    --h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               >!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        =         >  \         ]  {         |                                                C  A          n  l                                              J  W                                                          "         "  1#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            /         0  _         `                                               O         P  	         	  
         
                        ?         @  o         p                                    /         0  _         `                                               O         P                                                ?!         @!  o"         p"  #         D%  r&         t&  '         '  (         (  *         *  2+         4+  b,         d,  -         -  .         .  /         /  "1         $1  R2         T2  3         3  4         4  5         5  7         7  B8         D8  r9         t9  :         :  ;         ;  =         =  2>         4>  b?         d?  @         @  A         A  B         B  "D         $D  RE         TE  F         F  G         G  H          k  l          m  -n          [n  po          o  p          p  q         ir  s         s  t         t  u         u  (w         )w  Xx         Yx  y         y  z         z  {         {  }         }  H~         I~  x         y             ؁         ف           	  8         9  h         i             ȇ         ɇ             (         )  X         Y                                              H         I  x         y             ؔ         ٔ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               >!         J!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   ,      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-m-     m--    -;-x    ;-YJ-|    YJ-}-\    }-96-    96--h    -b-    b-n-	    n--h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               4!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                       !  A         B  b         c                       |                                    -  A          h  |                                              $  !          O  L                                              "         "  ;#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            (         *  R         T  |         ~                                   $         &  N	         P	  x
         z
                                             "  J         L  t         v                                              F         H  p         r                                              B         D  l         n               !         !  "          %  &          '  2(          `(  u)          )  *          *  +         K  L         L  N         N  1O         2O  [P         \P  Q         Q  R         R  S         S  U         U  -V         .V  WW         XW  X         X  Y         Y  Z         Z  [          \  )]         *]  S^         T^  }_         ~_  `         `  a         a  b         b  %d         &d  Oe         Pe  yf         zf  g         g  h         h  i         i  !k         "k  Kl         Ll  um         vm  n         n  o           6         8  `         b                        ޚ                    
  2         4  \         ^                        ڡ         ܡ             .         0  X         Z                        ֨         ب              *         ,  T         V  ~                      ү         ԯ             &         (  P         R  z         |             ζ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               M!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     8
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                 w        Yܯ--0     --     -c-     c-B-P    B-c-    c--    -'׳-p                             b      *      7   I      operation.0     sync             :            2            b         reason  ExchangeBarrier value   1       invocations     1                                     b      *      7   I      operation.0     sync             :            2            b         reason  ExchangeBarrier value   3       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   5       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   2       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-m-     m--    -;-x    ;-_J-|    _J-}-\    }-96-    96--h    -b-    b-n-	    n--h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               6!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             !         "  C         D  e         f                                                 
          2  E          m                                                )  '          T  R                                 !         !  !"         ""  @#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            )         *  S         T  }         ~                                   %         &  O	         P	  y
         z
                                   !         "  K         L  u         v                                              G         H  q         r                                              C         D  m         n               !         !  "          ),  >-          l-  .          .  /          /  1          51  J2         K  L         L  N         N  0O         2O  ZP         \P  Q         Q  R         R  S         S  U         U  ,V         .V  VW         XW  X         X  Y         Y  Z         Z  [          \  (]         *]  R^         T^  |_         ~_  `         `  a         a  b         b  $d         &d  Ne         Pe  xf         zf  g         g  h         h  i         i   k         "k  Jl         Ll  tm         vm  n         n  o           7         8  a         b                        ߚ           	         
  3         4  ]         ^                        ۡ         ܡ             /         0  Y         Z                        ר         ب             +         ,  U         V                        ӯ         ԯ             '         (  Q         R  {         |             ϶                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               Q!         R!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     7
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                 w        Yܯ--0     --     -c-     c-B-P    B-c-    c--    -'׳-p                             b      *      7   I      operation.0     sync             :            2            b         reason  ExchangeBarrier value   1       invocations     1                                     b      *      7   I      operation.0     sync             :            2            b         reason  ExchangeBarrier value   3       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   5       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   2       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,-C-     C--    -}5-    }5-D-    D-ew-l    ew-90-    90--x    -{b-	    {b--
    --
    -a-|    a-ճ-    ճ-!׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     f            !          N!  1          1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         )  "          R  K          {  t                                              /  3	          `	  d
          
                                   $         %  E         F  f         \  |         }                                                                  b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |        T$  v%         x%  &          :3  M4          }4  5          5  6          7  8          F8  Y9         vL  M         M  N         N  O         O  Q         Q  )R         *R  MS         NS  qT         rT  U         U  V         V  W         W  Y         Y  %Z         &Z  I[         J[  m\         n\  ]         ]  ^         ^  _         _  `         `  !b         "b  Ec         Fc  id         jd  e         e  f         f  g         g  h         h  j         j  Ak         Bk  el         fl  m         m  n         n  o         o  p         p  r         r  =s         >s  at         bt  u         u  v         v  w         w  x         x  z         z  9{         :{  ]|         ^|  }         }  ~         ~                                   5         6  Y         Z  }         ~             Ň         Ƈ                        1         2  U         V  y         z                                            
  ,         .  P         R  t         v                                              (         *  L         N  p         r                        ܠ         ޠ              $         &  H         J  l         n                        ب         ڨ                       "  D         F  h         j                        ԰                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               O!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    h
         i
             ;      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    h
         i
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -u,-`     u,-+-     +--    -;-    ;-5J-    5J-}-d    }-a-    a--p    -b-    b--
    --
    --t    --                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     g            !          >!  1          1  t=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                       !  A         B  b         c                                                           =  6          f  _                                              :  L          s           y                                     !         !  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         `  k                                              @  K         $  %         %  &         &  (         (  ')         ()  M*         N*  s+         t+  ,         ,  -         -  .         .  0         0  11         21  W2         X2  }3         ~3  4         4  5         5  6         6  8         8  ;9         <9  a:         b:  ;         ;  <         <  =         =  >         >  @          @  EA         FA  kB         lB  C         C  D         D  E         E  G         G  (H         *H  NI         PI  tJ         vJ  K         K  L         L  M         M  O         O  2P         4P  XQ         ZQ  ~R         R  S         S  T         T  U         U  W         W  <X         >X  bY         dY  Z         Z  [         [  \         \  ]         ]   _         "_  F`         H`  la         na  b         b  c         c  d         d  f         
f  0g         2g  Xh         Zh  i         i  j         j  k         k  l         l   n         "n  Ho         Jo  np         pp  q         q  r         r  s         s  u         u  ,v         .v  Rw         Tw  xx         zx  y         y  z         z  {         {  }         }  6~         8~  \         ^                        ΂         Ђ                        @         B  f         h                        ؊         ڊ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               Q!         R!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    h
         i
             :      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     /
                   
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            8     -,-h     ,-C-     C--    -}5-    }5-D-    D-ew-l    ew-90-    90--x    -ub-	    ub--
    --
    -a-|    a-ճ-    ճ-!׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     e            !          N!  1          1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         )  #          R  L          {  u                                              /  2	          `	  c
          
                                   $         %  E         F  f         \  |         }                                                                  b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |        T$  w%         x%  &          9  :          :  ;          <  "=          R=  e>          >  ?         vL  M         M  N         N  O         O  Q         Q  (R         *R  LS         NS  pT         rT  U         U  V         V  W         W   Y         Y  $Z         &Z  H[         J[  l\         n\  ]         ]  ^         ^  _         _  `         `   b         "b  Dc         Fc  hd         jd  e         e  f         f  g         g  h         h  j         j  @k         Bk  dl         fl  m         m  n         n  o         o  p         p  r         r  <s         >s  `t         bt  u         u  v         v  w         w  x         x  z         z  8{         :{  \|         ^|  }         }  ~         ~                                   4         6  X         Z  |         ~             ć         Ƈ                        0         2  T         V  x         z                                   	         
  -         .  Q         R  u         v                                              )         *  M         N  q         r                        ݠ         ޠ             %         &  I         J  m         n                        ٨         ڨ             !         "  E         F  i         j                        հ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               K!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    h
         i
             :      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    h
         i
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,-7-     7--    -;-    ;-;J-    ;J-}-d    }-a-    a--p    -b-    b--
    --
    --t    --                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     h            	!          B!  1          1  {=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                       !  A         B  b         c                                                           =  7          f  `                                              :  K          s           y                                      "         "  "#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         x                            !           "  +#          X#  c$         $  %         %  &         &   (         (  &)         ()  L*         N*  r+         t+  ,         ,  -         -  .         .  
0         0  01         21  V2         X2  |3         ~3  4         4  5         5  6         6  8         8  :9         <9  `:         b:  ;         ;  <         <  =         =  >         >  @          @  DA         FA  jB         lB  C         C  D         D  E         E  G         G  )H         *H  OI         PI  uJ         vJ  K         K  L         L  M         M  O         O  3P         4P  YQ         ZQ  R         R  S         S  T         T  U         U  W         W  =X         >X  cY         dY  Z         Z  [         [  \         \  ]         ]  !_         "_  G`         H`  ma         na  b         b  c         c  d         d  	f         
f  1g         2g  Yh         Zh  i         i  j         j  k         k  l         l  !n         "n  Io         Jo  op         pp  q         q  r         r  s         s  u         u  -v         .v  Sw         Tw  yx         zx  y         y  z         z  {         {  }         }  7~         8~  ]         ^                        ς         Ђ                        A         B  g         h                        ي         ڊ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               Q!         R!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    h
         i
             ;      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     0
                   
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,-U-     U--    -;-x    ;-;J-|    ;J-}-\    }-c3-    c3--h    -b-    b-m-	    m--h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     h            !          T!  1          1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                       !  A         B  b         c                       r  z                                    %          V  ^                                              
            3  ,         y                                     !         !  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                      '  G         N  n         u                                   
           1	         8	  X
         _
                                                       "  B         I  i         p                                              ,         3  S         Z  z                                                       =          D   d!         k!  "          <  $=          T=  ^>          >  ?          ?  @          A  B         K  L         L  N         N  5O         >O  cP         lP  Q         Q  R         R  S         S  U         $U  IV         RV  wW         W  X         X  Y         Y  Z         [  '\         0\  S]         \]  ^         ^  _         _  `         `  b         b  /c         8c  Yd         bd  e         e  f         f  g         g  i         
i  +j         4j  Uk         ^k  l         l  m         m  n         ȑ                        0         4  T         X  x         |                      ę                        ,         0  P         T  t         x                                              (         ,  L         P  p         t                        ܪ                       $         (  H         L  l         p                        ز         ܲ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               V!         Z!  2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     ,
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            8     -,-h     ,-a-     a--    -4-    4-uC-    uC-v-l    v---    ---x    --b-	    -b--
    --
    -m-|    m-ճ-    ճ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               1!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        9  K          r                                                0            C  .	          V	  A
          i
  T         |                                                                                            5         6  V                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                        %            J  B          o  g                     V"  {#         }#  $         H  I         I  J         J  K         K  M         M  EN         GN  mO         oO  P         P  Q         Q  R         R  T         T  5U         7U  ]V         _V  W         W  X         X  Y         Y  Z         Z  %\         '\  M]         O]  u^         w^  _         _  `         `  a         a  c         c  =d         ?d  ee         ge  f         f  g         g  h         h  j         j  -k         /k  Ul         Wl  }m         m  n         n  o         o  p         p  r         r  Es         Gs  mt         ot  u         u  v         v  w         w  y         y  5z         7z  ]{         _{  |         |  }         }  ~         ~             %         '  M         O  u         w             Ņ         ǅ                        =         ?  e         g                        ݍ         ߍ             ,         .  S         U  {         }             ˔         ͔                        C         E  k         m                                              3         5  [         ]                        ӣ         գ             #         %  K         M  s         u             ê         Ū                        ;                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     g            	!          D!  1          1  }=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    h
         i
             :      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    j
         k
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            0     -' -X     ' -հ-     հ-۰-    ۰-4-p    4-W@-T    W@-h-4    h-c3-    c3--8    -b-    b-m-	    m--8
    --
                   =         b      J      7   U   7   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     i            !          T!  !1                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 64 })  outputElements  5184    input0Elements  31104   input1Elements  24576   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     4                           #         b      J      7   M     operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       @                       !  A         B  b         c            (	  /
          a
  h                                              5  /          ^  X                                                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     4                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  1728    input0Elements  1728    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       x            !         '  H         N  o         u                                              2	         8	  Y
         _
                                                       "  C         I  j         p                                              -         3  T         Z  {                                                       >          D   e!         k!  "          <B  FC          vC  D          D  E          E  F         K  L         L  N         N  4O         >O  bP         lP  Q         Q  R         R  S         S  U         $U  HV         RV  vW         W  X         X  Y         Y  Z         [  &\         0\  R]         \]  ~^         ^  _         _  `         `  b         b  .c         8c  Xd         bd  e         e  f         f  g         g   i         
i  *j         4j  Tk         ^k  ~l         l  m         m  n         ȑ                        1         4  U         X  y         |                      ę             	           -         0  Q         T  u         x                                              )         ,  M         P  q         t                        ݪ                      %         (  I         L  m         p                        ٲ         ܲ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               T!         Z!  2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     +
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            8     -,-h     ,-g-     g--    -4-    4-{C-    {C-v-l    v---    ---x    --b-	    -b--
    --
    -m-|    m-ճ-    ճ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               3!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        9  J          r                                                0            C  -	          V	  @
          i
  S         |                                              %                                            ?         @  `                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                             (	   
          M
  E         V"  |#         }#  $         H  I         I  J         J  K         K  M         M  FN         GN  nO         oO  P         P  Q         Q  R         R  T         T  6U         7U  ^V         _V  W         W  X         X  Y         Y  Z         Z  &\         '\  N]         O]  v^         w^  _         _  `         `  a         a  c         c  >d         ?d  fe         ge  f         f  g         g  h         h  j         j  .k         /k  Vl         Wl  ~m         m  n         n  o         o  p         p  r         r  Fs         Gs  nt         ot  u         u  v         v  w         w  y         y  6z         7z  ^{         _{  |         |  }         }  ~         ~             &         '  N         O  v         w             ƅ         ǅ                        >         ?  f         g                        ލ         ߍ             -         .  T         U  |         }             ̔         ͔                        D         E  l         m                                              4         5  \         ]                        ԣ         գ             $         %  L         M  t         u             Ī         Ū                        <                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     h            
!          D!  1          1  ~=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    h
         i
             ;      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    k
         l
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            0     -,-X     ,--     -#-    #-5-x    5-#D-|    #D-qw-\    qw--    --h    -Wb-    Wb-wx-	    wx--h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ?!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         )  "          R  K          {  t                                              /  3	          `	  d
          
                                   $         %  E         F  f         \  |         }                                                                  b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                     @  0          `  P            p         $  %         %  '         '  8(         :(  j)         l)  *         *  +         +   -         -  2.         4.  d/         f/  0         0  1         1  2         2  ,4         .4  ^5         `5  6         6  7         7  8         8  &:         (:  X;         Z;  <         <  =         =  >         >   @         "@  RA         TA  B         B  C         C  D         D  F         F  LG         NG  ~H         H  I         I  K         K  BL         DL  tM         vM  N         N  O         O  
Q         Q  <R         >R  nS         pS  T         T  U         U  W         W  7X         8X  iY         jY  Z         Z  [         [  \          ]  1^         2^  c_         d_  `         `  a         a  b         b  +d         ,d  ]e         ^e  f         f  g         g  h         h  %j         &j  Wk         Xk  l         l  m         m  n         n  p          p  Qq         Rq  r         r  s         s  t         t  v         v  Kw         Lw  }x         ~x  y         y  z         z  |         |  E}         F}  w~         x~             ۀ         ܀             ?         @  q         r             Ն         ֆ             9         :  k         l             ό         Ќ             3         d                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     l            !          X!  1          2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 o
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-g-     g--    -;-x    ;-AJ-|    AJ-}-\    }-.-    .--h    -9b-    9b-n-	    n--h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               2!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                       !  A         B  b         c                                                           =  6          f  _                                              :  L          s           y                                     !         !  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            (         *  R         T  |         ~                                   $         &  N	         P	  x
         z
                                             "  J         L  t         v                                              F         H  p         r                                              B         D  l         n               !          /  0          0  1          1  2          3  4          @4  H5         @G  iH         jH  I         I  J         J  K         K  "M         $M  LN         NN  vO         xO  P         P  Q         Q  R         R  T          T  HU         JU  rV         tV  W         W  X         X  Y         Y  [         [  D\         F\  n]         p]  ^         ^  _         _  `         `  b         b  @c         Bc  jd         ld  e         e  f         f  g         g  i         i  <j         >j  fk         hk  l         l  m           ͏         Ώ             !         "  K         L  u         v             ɖ         ʖ                        G         H  q         r             ŝ         Ɲ                        C         D  m         n                      ¤                        ?         @  i         j                                              ;                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     g            !          J!  1          1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     O
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-5-x    5-#D-|    #D-qw-\    qw--    --h    -Kb-    Kb-wx-	    wx--h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               A!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         )  #          R  L          {  u                                              /  2	          `	  c
          
                                   $         %  E         F  f         \  |         }                                                                  b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                              	  	           
           $  %         %  '         '  9(         :(  k)         l)  *         *  +         +  -         -  3.         4.  e/         f/  0         0  1         1  2         2  -4         .4  _5         `5  6         6  7         7  8         8  ':         (:  Y;         Z;  <         <  =         =  >         >  !@         "@  SA         TA  B         B  C         C  D         D  F         F  MG         NG  H         H  I         I  K         K  CL         DL  uM         vM  N         N  O         O  Q         Q  =R         >R  oS         pS  T         T  U         U  W         W  6X         8X  hY         jY  Z         Z  [         [  \          ]  0^         2^  b_         d_  `         `  a         a  b         b  *d         ,d  \e         ^e  f         f  g         g  h         h  $j         &j  Vk         Xk  l         l  m         m  n         n  p          p  Pq         Rq  r         r  s         s  t         t  v         v  Jw         Lw  |x         ~x  y         y  z         z  |         |  D}         F}  v~         x~             ڀ         ܀             >         @  p         r             Ԇ         ֆ             8         :  j         l             Ό         Ќ              2         d                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     k            !          T!  1          1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 o
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-g-     g--    -;-x    ;-AJ-|    AJ-}-\    }-.-    .--h    -?b-    ?b-n-	    n--h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               2!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                       !  A         B  b         c                                                           =  7          f  `                                              :  K          s           y                                      "         "  "#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            )         *  S         T  }         ~                                   %         &  O	         P	  y
         z
                                   !         "  K         L  u         v                                              G         H  q         r                                              C         D  m         n               !          p5  x6          6  7          7  8           9  :          0:  8;         @G  hH         jH  I         I  J         J  K         K  #M         $M  MN         NN  wO         xO  P         P  Q         Q  R         R  T          T  IU         JU  sV         tV  W         W  X         X  Y         Y  [         [  E\         F\  o]         p]  ^         ^  _         _  `         `  b         b  Ac         Bc  kd         ld  e         e  f         f  g         g  i         i  =j         >j  gk         hk  l         l  m           ̏         Ώ                       "  J         L  t         v             Ȗ         ʖ                        F         H  p         r             ĝ         Ɲ                        B         D  l         n                      ¤                        >         @  h         j                                              :                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     h            !          N!  1          1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     N
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-;-x    ;-I-|    I-|-\    |-1-    1-w-h    w-b-    b-!-	    !--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               G!         \!  1         
2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P           =         >  ^         _  	         	  
         
                                                       	  )          *            >            R  *          f  >          z  R                                                9   A!          r!  z"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            .         0  ^         `                                               N         P  ~	         	  
         
                        >         @  n         p                                    .         0  ^         `                                               N         P  ~                                              >!         @!  n"         p"  #         #  $         $  %          &  .'         0'  ^(         `(  )         )  *         *  +         +  -          -  N.         P.  ~/         /  0         0  1         1  3         3  >4         @4  n5         p5  6         6  7         7  8          9  .:         0:  ^;         `;  <         <  =         =  >         >  @          @  NA         PA  ~B         B  C         C  D         D  F         F  >G         @G  nH         pH  I         I  J          (y  6z          kz  y{          {  |          |  }          4~  B         d                      Ĕ             "         $  R         T                                              B         D  r         t             ҡ         ԡ             2         4  b         d             §         ħ             "         $  R         T                                              B         D  r                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     [                       8!  1          1  _=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   R      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-m-     m--    -4-x    4-{C-|    {C-v-\    v--    --h    -ib-    ib-n-	    n--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               4!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        9  K          r                                                0            C  .	          V	  A
          i
  T         |                                                                                            5         6  V                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            *         ,  V         X                                              2         4  ^	         `	  
         
                                   :         <  f         h                                              B         D  n         p                                               J         L  v         x                          !         !  &#          d<  |=          =  >          >  ?          $@  <A          dA  |B         HI  rJ         tJ  K         K  L         L  M         M  "O         $O  NP         PP  zQ         |Q  R         R  S         S  T          U  *V         ,V  VW         XW  X         X  Y         Y  Z         Z  \         \  2]         4]  ^^         `^  _         _  `         `  a         a  c         c  :d         <d  fe         he  f         f  g         g  h         h  j         j  Bk         Dk  nl         pl  m         m  n         n  o         o  q          q  Jr         Lr  vs         xs  t         t  u         u  v         v  &x         (x  Ry         Ty  ~z         z  {         {  |         |  ~         ~  .         0  Z         \                        ރ           
           6         8  b         d                                              >         @  j         l                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               8!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     O
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-;-x    ;-I-|    I-|-\    |-1-    1--h    -!b-    !b-!-	    !--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               E!         X!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P           =         >  ^         _  	         	  
         
                                                         .          /            D            Y  1          n  F            [                                  
             C   J!          |!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            /         0  _         `                                               O         P  	         	  
         
                        ?         @  o         p                                    /         0  _         `                                               O         P                                                ?!         @!  o"         p"  #         #  $         $  %          &  /'         0'  _(         `(  )         )  *         *  +         +  -          -  O.         P.  /         /  0         0  1         1  3         3  ?4         @4  o5         p5  6         6  7         7  8          9  /:         0:  _;         `;  <         <  =         =  >         >  @          @  OA         PA  B         B  C         C  D         D  F         F  ?G         @G  oH         pH  I         I  J          w              ȁ                      @  N                     d             Ô         Ĕ             #         $  S         T                                              C         D  s         t             ӡ         ԡ             3         4  c         d             ç         ħ             #         $  S         T                                              C         D  s                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     \                       <!  1          1  f=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   Q      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-m-     m--    -4-x    4-C-|    C-v-\    v-	-    	--h    -cb-    cb-n-	    n--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               4!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        9  J          r                                                0            C  -	          V	  @
          i
  S         |                                              %                                            ?         @  `                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            +         ,  W         X                                              3         4  _	         `	  
         
                                   ;         <  g         h                                              C         D  o         p                                               K         L  w         x                          !         !  '#          B  C          C  D          $E  <F          dF  |G          G  H         HI  sJ         tJ  K         K  L         L  M         M  #O         $O  OP         PP  {Q         |Q  R         R  S         S  T          U  +V         ,V  WW         XW  X         X  Y         Y  Z         Z  \         \  3]         4]  _^         `^  _         _  `         `  a         a  c         c  ;d         <d  ge         he  f         f  g         g  h         h  j         j  Ck         Dk  ol         pl  m         m  n         n  o         o  q          q  Kr         Lr  ws         xs  t         t  u         u  v         v  'x         (x  Sy         Ty  z         z  {         {  |         |  ~         ~  /         0  [         \                        ߃                      7         8  c         d                                              ?         @  k         l                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               4!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     N
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-;-x    ;-GJ-|    GJ-}-\    }--    --h    -9b-    9b-'-	    '--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ;!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                       !  A         B  b         c                       r  z                                    %          V  ^                                              
            3  ,         y                                     !         !  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ,         .  Z         \                                              @         B  n	         p	  
         
                        &         (  T         V                                              :         <  h         j                                             "  N         P  |         ~                        !         !  4"         6"  b#         d#  $         $  %         %  &         &  (         (  H)         J)  v*         x*  +         +  ,         ,   .         .  ./         0/  \0         ^0  1         1  2         2  3         3  5         5  B6         D6  p7         r7  8         8  9         9  :         :  (<         *<  V=         X=  >         >  ?         ?  @         @  B         B  <C         >C  jD         lD  E         E  F          X`  da          a  b          b  c          d  e          Pe  \f         l  m         m  o          o  Mp         Np  {q         |q  r         r  s         s  u         u  3v         4v  aw         bw  x         x  y         y  z         z  |         |  G}         H}  u~         v~             р         Ҁ              -         .  [         \                                              A         B  o         p             ˍ         ̍             '                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               !         .!  1         1  }=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   R      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     --    -/;-x    /;-I-|    I-}-\    }-g-    g--h    -]b-    ]b-x-	    x--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               <!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 /	         0	  P
         Q
  q         g                                                                    =  7          n  h                                              *  #          S  L           |   u!          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            -         .  [         \                                              A         B  o	         p	  
         
                        '         (  U         V                                              ;         <  i         j                                   !         "  O         P  }         ~                        !         !  5"         6"  c#         d#  $         $  %         %  &         &  (         (  H)         J)  v*         x*  +         +  ,         ,   .         .  ./         0/  \0         ^0  1         1  2         2  3         3  5         5  B6         D6  p7         r7  8         8  9         9  :         :  (<         *<  V=         X=  >         >  ?         ?  @         @  B         B  <C         >C  jD         lD  E         E  F         F  G         G  "I         $I  PJ         RJ  ~K         K  L         L  M         M  O         
O  6P         8P  dQ         fQ  R         R  S         S  T         T  V         V  JW         LW  xX         zX  Y         Y  Z         Z  \         \  0]         2]  ^^         `^  _         _  `         `  a         a  c         c  Dd         Fd  re         te  f         f  g         g  h          h  i          "j  k          Fk  2l          jl  Vm          m  zn         ʐ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               1!         B!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     --    -;-x    ;-GJ-|    GJ-}-\    }--    --h    -?b-    ?b-!-	    !--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               9!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                       !  A         B  b         c                       r  y                                    $          V  ]                                              
            3  -         y                                     !         !  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            -         .  [         \                                              A         B  o	         p	  
         
                        '         (  U         V                                              ;         <  i         j                                   !         "  O         P  }         ~                        !         !  5"         6"  c#         d#  $         $  %         %  &         &  (         (  I)         J)  w*         x*  +         +  ,         ,  .         .  //         0/  ]0         ^0  1         1  2         2  3         3  5         5  C6         D6  q7         r7  8         8  9         9  :         :  )<         *<  W=         X=  >         >  ?         ?  @         @  B         B  =C         >C  kD         lD  E         E  F          f  g          g  h          
i  j          Hj  Tk          k  l         l  m         m  o          o  Lp         Np  zq         |q  r         r  s         s  u         u  2v         4v  `w         bw  x         x  y         y  z         z  |         |  F}         H}  t~         v~             Ѐ         Ҁ              ,         .  Z         \                                              @         B  n         p             ʍ         ̍             &                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               #!         2!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   Q      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-y-     y--    -/;-x    /;-I-|    I-}-\    }-m-    m--h    -cb-    cb-x-	    x--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               8!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 /	         0	  P
         Q
  q         g                                                                    =  6          n  g                                              *  $          S  M           |   v!          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ,         .  Z         \                                              @         B  n	         p	  
         
                        &         (  T         V                                              :         <  h         j                                             "  N         P  |         ~                        !         !  4"         6"  b#         d#  $         $  %         %  &         &  (         (  I)         J)  w*         x*  +         +  ,         ,  .         .  //         0/  ]0         ^0  1         1  2         2  3         3  5         5  C6         D6  q7         r7  8         8  9         9  :         :  )<         *<  W=         X=  >         >  ?         ?  @         @  B         B  =C         >C  kD         lD  E         E  F         F  G         G  #I         $I  QJ         RJ  K         K  L         L  M         M  	O         
O  7P         8P  eQ         fQ  R         R  S         S  T         T  V         V  KW         LW  yX         zX  Y         Y  Z         Z  \         \  1]         2]  _^         `^  _         _  `         `  a         a  c         c  Ed         Fd  se         te  f         f  g         g  h          n  o          o  p          p  q          r  
s          Bs  .t         ʐ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               3!         F!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     --    -;-x    ;-I-|    I-|-\    |-k
-    k
--P    -ob-    ob-c-	    c-a-D
    a-3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               8!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P           =         >  ^         _  	         	  
         
                                                       	  )          *            >            R  *          f  >          z  R                                                9   A!          r!  z"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            *         ,  V         X                                              2         4  ^	         `	  
         
                                   :         <  f         h                                              B         D  n         p                                               J         L  v         x                          !         !  &#         (#  R$         T$  ~%         %  &         &  '         '  )         )  .*         0*  Z+         \+  ,         ,  -         -  .         .  
0         0  61         81  b2         d2  3         3  4         4  5         5  7         7  >8         @8  j9         l9  :         :  ;         ;  <         <  >         >  F?         H?  r@         t@  A         A  B         B  C          C  D          E  F          6F  #G          UG  BH          tH  aI         h  (j         *j  Tk         Vk  l         l  m         m  n         n  p         p  0q         2q  \r         ^r  s         s  t         t  u         u  w         w  8x         :x  dy         fy  z         z  {         {  |         |  ~         ~  @         B  l         n                      Ă                        C         E  o         q             ǉ         ɉ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               4!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-y-     y--    -4-x    4-C-|    C-v-\    v-c0-    c0--h    -Qb-    Qb-x-	    x--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               6!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        9  K          r                                                0            C  .	          V	  A
          i
  T         |                                                                                            5         6  V                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         @  5          `  U            u                                 %  	'         
'  9(         :(  i)         j)  *         *  +         +  ,         ,  ).         *.  Y/         Z/  0         0  1         1  2         2  4         4  I5         J5  y6         z6  7         7  8         8  	:         
:  9;         :;  i<         j<  =         =  >         >  ?         ?  )A         *A  YB         ZB  C         C  D         D  E         E  G         G  IH         JH  yI         h  i         i  *k         ,k  Zl         \l  m         m  n         n  o         o  q         q  Jr         Lr  zs         |s  t         t  u         u  
w         w  :x         <x  jy         ly  z         z  {         {  |         |  *~         ,~  Z         \                                              J         L  z         |             ڈ         ܈  
           :         <  j         P  ~                      ސ                      >         @  n         p             Ζ         Ж              .         0  ^         `                                               N         P  ~                      ޣ                      >         @  n         p             Ω         Щ              .         0  ^         `                                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               *!         4!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-y-     y--    -;-x    ;-I-|    I-|-\    |-	-    	--P    -ub-    ub-c-	    c-a-D
    a-3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               7!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P           =         >  ^         _  	         	  
         
                                                         .          /            D            Y  1          n  F            [                                  
             C   J!          |!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            +         ,  W         X                                              3         4  _	         `	  
         
                                   ;         <  g         h                                              C         D  o         p                                               K         L  w         x                          !         !  '#         (#  S$         T$  %         %  &         &  '         '  )         )  /*         0*  [+         \+  ,         ,  -         -  .         .  0         0  71         81  c2         d2  3         3  4         4  5         5  7         7  ?8         @8  k9         l9  :         :  ;         ;  <         <  >         >  G?         H?  s@         t@  A         A  B         B  C         h  )j         *j  Uk         Vk  l         l  m         m  n         n  p         p  1q         2q  ]r         ^r  s         s  t         t  u         u  w         w  9x         :x  ey         fy  z         z  {         {  |         |  ~         ~  A         B  m         n             Â         Ă                        D         E  p         q             ȉ         ɉ            (            G  4          f  S            r                               >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               6!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-m-     m--    -4-x    4-C-|    C-v-\    v-i0-    i0--h    -Wb-    Wb-x-	    x--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               2!         @!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        9  J          r                                                0            C  -	          V	  @
          i
  S         |                                              %                                            ?         @  `                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                               @  5          `  U         %  '         
'  8(         :(  h)         j)  *         *  +         +  ,         ,  (.         *.  X/         Z/  0         0  1         1  2         2  4         4  H5         J5  x6         z6  7         7  8         8  :         
:  8;         :;  h<         j<  =         =  >         >  ?         ?  (A         *A  XB         ZB  C         C  D         D  E         E  G         G  HH         JH  xI         h  i         i  +k         ,k  [l         \l  m         m  n         n  o         o  q         q  Kr         Lr  {s         |s  t         t  u         u  w         w  ;x         <x  ky         ly  z         z  {         {  |         |  +~         ,~  [         \                                              K         L  {         |             ۈ         ܈             ;         <  k         P                        ߐ                      ?         @  o         p             ϖ         Ж              /         0  _         `                                               O         P                        ߣ                      ?         @  o         p             ϩ         Щ              /         0  _         `                                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ,!         8!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,-y-     y--    -;-    ;-MJ-    MJ-}-l    }-G-    G--P    --    -b-h	    b--p
    -#-
    #--    -9׳-L                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               5!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                       !  A         B  b         c                       r  z                                    %          V  ^                                              
            3  ,         y                                     !         !  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        %  1&         2&  ['         \'  (         (  )         )  *         *  ,         ,  --         .-  W.         X.  /         /  0         0  1         1  2          3  )4         *4  S5         T5  }6         ~6  7         7  8         8  9         9  %;         &;  O<         P<  y=         z=  >         >  ?         ?  @         @  !B         "B  KC         LC  uD         vD  E         E  F         F  G         G  I         I  FJ         HJ  pK         rK  L         L  M         M  N         N  P         P  BQ         DQ  lR         nR  S         S  T         T  U         U  W         W  >X         @X  hY         jY  Z         Z  [         [  \         \  ^         ^  :_         <_  d`         f`  a         a  b         b  c         c  e         e  6f         8f  `g         bg  h         h  i         i  j         j  l         
l  2m         4m  \n         ^n  o         o  p         p  q         q  s         s  .t         0t  Xu         Zu  v          v  w          w  x          x  y          z  z          ,{  )|         V|  ~}         }  ~         ~                        &         (  P         R  z         |             Ά         І             "         $  L         N  v         x             ʍ                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               A!         Z!  1         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    h
         i
             ;                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     H
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-g-     g--    -5;-x    5;-I-|    I-}-\    }-C-    C--h    -cb-    cb--	    --h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               -!         8!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 /	         0	  P
         Q
  q         g                                                                    =  7          n  h                                              *  #          S  L           |   u!          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            1         2  c         d                                   +         ,  ]         ^  	         	  
         
             %         &  W         X                                               Q         R                                              K         L  }         ~                                   E          F   w!         x!  "         "  #          =  >          >   @          0@  <A          lA  xB          B  C         G  H         H  I         I  -K         .K  _L         `L  M         M  N         N  O         O  'Q         (Q  YR         ZR  S         S  T         T  U         U  !W         "W  SX         TX  Y         Y  Z         Z  [         [  ]         ]  M^         N^  _         _  `         `  a         a  c         c  Gd         Hd  ye         ze  f         f  g         g  i         i  Aj         Bj  sk         tk  l         l  m         Pp  q         q  r         r  s         s  u         u  Hv         Jv  zw         |w  x         x  y         y  {         {  B|         D|  t}         v}  ~         ~             
           <         >  n         p             ҅         ԅ             6         8  h         j             ̋         ΋              0         2  b         d             Ƒ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               -!         @!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   #      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,-y-     y--    -;-    ;-MJ-    MJ-}-l    }-G-    G--P    --    -b-h	    b--p
    -#-
    #--    -9׳-L                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               5!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                       !  A         B  b         c                       r  y                                    $          V  ]                                              
            3  -         y                                     !         !  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        %  0&         2&  Z'         \'  (         (  )         )  *         *  ,         ,  ,-         .-  V.         X.  /         /  0         0  1         1  2          3  (4         *4  R5         T5  |6         ~6  7         7  8         8  9         9  $;         &;  N<         P<  x=         z=  >         >  ?         ?  @         @   B         "B  JC         LC  tD         vD  E         E  F         F  G         G  I         I  GJ         HJ  qK         rK  L         L  M         M  N         N  P         P  CQ         DQ  mR         nR  S         S  T         T  U         U  W         W  ?X         @X  iY         jY  Z         Z  [         [  \         \  ^         ^  ;_         <_  e`         f`  a         a  b         b  c         c  e         e  7f         8f  ag         bg  h         h  i         i  j         j  	l         
l  3m         4m  ]n         ^n  o         o  p         p  q         q  s         s  /t         0t  Yu         Zu  v         v  w         w  x         x  z         z  +{         ,{  U|          V|  S}          }  }~          ~              р                     (  Q         R  {         |             φ         І             #         $  M         N  w         x             ˍ                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               A!         Z!  1         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    h
         i
             :                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     I
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-a-     a--    -5;-x    5;-I-|    I-}-\    }-C-    C--h    -Wb-    Wb--	    --h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               *!         6!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 /	         0	  P
         Q
  q         g                                                                    =  6          n  g                                              *  $          S  M           |   v!          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            0         2  b         d                                   *         ,  \         ^  	         	  
         
             $         &  V         X                                               P         R                                              J         L  |         ~                                   D          F   v!         x!  "         "  #          C  D           E  ,F          \F  hG         G  H         H  I         I  ,K         .K  ^L         `L  M         M  N         N  O         O  &Q         (Q  XR         ZR  S         S  T         T  U         U   W         "W  RX         TX  Y         Y  Z         Z  [         [  ]         ]  L^         N^  ~_         _  `         `  a         a  c         c  Fd         Hd  xe         ze  f         f  g         g  i         i  @j         Bj  rk         tk  l         l  m          m  n          o   p         Pp  q         q  r         r  s         s  u         u  Iv         Jv  {w         |w  x         x  y         y  {         {  C|         D|  u}         v}  ~         ~                        =         >  o         p             Ӆ         ԅ             7         8  i         j             ͋         ΋              1         2  c         d             Ǒ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               )!         <!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   $      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     -#-    #-5-    5-YD-    YD-w-l    w-12-    12--x    -{b-	    {b--
    -)-
    )-y-|    y-ճ-    ճ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               7!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         )  "          R  K          {  t                                              /  0	          `	  a
          
                                   $         %  E         F  f         \                                              #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                   	          6  C          p  }         $  %         K  L         L  N         N  EO         FO  mP         nP  Q         Q  R         R  S         S  U         U  5V         6V  ]W         ^W  X         X  Y         Y  Z         Z  [         [  %]         &]  M^         N^  u_         v_  `         `  a         a  b         b  d         d  =e         >e  ef         ff  g         g  h         h  i         i  k         k  -l         .l  Um         Vm  }n         ~n  o         o  p         p  q         q  s         s  Et         Ft  mu         nu  v         v  w         w  x         x  z         z  5{         6{  ]|         ^|  }         }  ~         ~                        %         &  M         N  u         v             ņ         Ɔ                        =         >  e         f                        ݎ         ގ             -         .  T         V  |         ~             ̕         Ε                        D         F  l         n                                              4         6  \         ^                        Ԥ         ֤             $         &  L         N  t         v             ī         ƫ                        <         >  d         f                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               8!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    h
         i
             ;      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    h
         i
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     `
            4     -,-`     ,--     -)-    )-<-    <-J-    J-}-d    }--    --p    -{b-    {b-!-
    !-/-
    /-۰-t    ۰-9׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               >!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                       !  A         B  b         c                                                           =  6          f  _                                              :  G          s                                   !          !  K"         L"  w#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       4  ;          h  o                                 $  8%         :%  [&         \&  }'         ~'  (         (  )         )  *         *  ,         ,  '-         (-  I.         J.  k/         l/  0         0  1         1  2         2  3         3  5         5  76         86  Y7         Z7  {8         |8  9         9  :         :  ;         ;  =         =  %>         &>  G?         H?  i@         j@  A         A  B         B  C         C  D         D  F         F  5G         6G  VH         XH  xI         zI  J         J  K         K  L         L  "N         $N  UO         WO  P         P  Q         Q  R         R  !T         #T  TU         VU  V         V  W         W  X         X  'Z         (Z  ][         ^[  \         \  ]         ]  ^          _  5`         6`  ka         la  b         b  c         c  e         e  ?f         @f  sg         tg  h         h  i         i  k         k  Cl         Dl  wm         xm  n         n  o         o   q         q  "r         $r  Ds         Fs  ft         ht  u         u  v         v  w         w  x         x  z         z  2{         4{  T|         V|  v}         x}  ~         ~             ܀         ހ                        "  B         D  d         f                        ʈ         ̈                        0                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               6!         J!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    h
         i
             E      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     /
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     -#-    #-5-    5-SD-    SD-w-l    w-72-    72--x    -ub-	    ub--
    -)-
    )-y-|    y-ճ-    ճ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               7!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         )  #          R  L          {  u                                               /  1	          `	  b
          
                                   $         %  E         F  f         \                                                                            b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                  !  +"          X"  e#          #  $         $  %         K  L         L  N         N  DO         FO  lP         nP  Q         Q  R         R  S         S  U         U  4V         6V  \W         ^W  X         X  Y         Y  Z         Z  [         [  $]         &]  L^         N^  t_         v_  `         `  a         a  b         b  d         d  <e         >e  df         ff  g         g  h         h  i         i  k         k  ,l         .l  Tm         Vm  |n         ~n  o         o  p         p  q         q  s         s  Dt         Ft  lu         nu  v         v  w         w  x         x  z         z  4{         6{  \|         ^|  }         }  ~         ~                        $         &  L         N  t         v             Ć         Ɔ                        <         >  d         f                        ܎         ގ             ,         .  U         V  }         ~             ͕         Ε                        E         F  m         n                                              5         6  ]         ^                        դ         ֤             %         &  M         N  u         v             ū         ƫ                        =         >  e         f                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               6!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    h
         i
             :      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    h
         i
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     _
            4     -,-`     ,--     -/-    /-<-    <-J-    J-}-d    }--    --p    -ub-    ub-'-
    '-5-
    5-۰-t    ۰-9׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               @!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                       !  A         B  b         c                                                           =  7          f  `                                              :  F          s                                   !         !  J"         K"  w#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                     8  ?          l  s	          	  
          
           $  9%         :%  Z&         \&  |'         ~'  (         (  )         )  *         *  ,         ,  &-         (-  H.         J.  j/         l/  0         0  1         1  2         2  3         3  5         5  66         86  X7         Z7  z8         |8  9         9  :         :  ;         ;  =         =  $>         &>  F?         H?  h@         j@  A         A  B         B  C         C  D         D  F         F  4G         6G  WH         XH  yI         zI  J         J  K         K  L         L  #N         $N  VO         WO  P         P  Q         Q  R         R  "T         #T  UU         VU  V         V  W         W  X         X  &Z         (Z  \[         ^[  \         \  ]         ]  ^          _  4`         6`  ja         la  b         b  c         c  
e         e  >f         @f  rg         tg  h         h  i         i  k         k  Bl         Dl  vm         xm  n         n  o         o  q         q  #r         $r  Es         Fs  gt         ht  u         u  v         v  w         w  x         x  z         z  3{         4{  U|         V|  w}         x}  ~         ~             ݀         ހ              !         "  C         D  e         f                        ˈ         ̈                        1                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               4!         J!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    h
         i
             F      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     0
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -A-    A-;-x    ;-YJ-|    YJ-}-\    }-4-    4--h    -b-    b-	n-	    	n--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               K!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                       !  A         B  b         c                       r  o                                              V  S                                              
            3  ,         y                                     !         !  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            $         &  J         L  p         r                                            
  .	         0	  T
         V
  z         |                                              8         :  ^         `                                                         B         D  h         j                                               &          (   L!         N!  r"          #  $$          T$  Z%          %  &          &  '          '  (         K  L         L  M          N  %O         &O  KP         LP  qQ         rQ  R         R  S         S  T         T  	V         
V  /W         0W  UX         VX  {Y         |Y  Z         Z  [         [  \         \  ^         ^  9_         :_  _`         ``  a         a  b         b  c         c  d         d  f         f  Cg         Dg  ih         jh  i         i  j         j  k         k  m         m  'n         (n  Mo           ?         @  e         f                        י         ؙ             #         $  I         J  o         p                                              -         .  S         T  y         z             Ũ         ƨ                        7         8  ]         ^                        ϰ         а                        A         B  g                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               L!         R!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     !
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    <     -,-p     ,--     -A-    A-#5-    #5-C-    C-w-t    w-5-     5-5=-X    5=--    -b-p	    b-!-x
    !-/-
    /--    -ճ-\    ճ--<                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               I!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P              
          9  C          r  |                                              0            C  	          V	  1
          i
  D         |                                                                                 <         =  ]         ^  ~                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        8"  Y#         [#  |$         ~$  %         %  &         &  '         '  )         )  0*         2*  T+         V+  x,         z,  -         -  .         .  /         /  1         
1  ,2         .2  P3         R3  t4         v4  5         5  6         6  7         7  9         9  (:         *:  L;         N;  p<         r<  =         =  >         >  ?         ?   A         A  $B         &B  HC         JC  lD         nD  E         E  F         F  G         -l  Nm         Pm  qn         sn  o         o  p         p  q         q  s         s  %t          't  u          Ku  Bv          ov  fw          w  x          x  y         y  z         z  !|         #|  E}         G}  i~         k~                        Ձ         ׁ                        A         C  e         g                        щ         Ӊ           ړ                       "  D         F  h         j                        ԛ         ֛                        @         B  d         f                        У         ң                        <         >  `         b                        ̫         Ϋ                        8         :  \         ^                        ȳ         ʳ                                           b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               H!         L!  1         1  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    h
         i
             :                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    h
         i
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     O
            0     -,-X     ,--     -A-    A-;-x    ;-YJ-|    YJ-}-\    }-4-    4--h    -b-    b-	n-	    	n--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               K!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                       !  A         B  b         c                       r  n                                              V  R                                              
            3  -         y                                     !         !  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            %         &  K         L  q         r                                   	         
  /	         0	  U
         V
  {         |                                              9         :  _         `                                                         C         D  i         j                                              '          (   M!         N!  s"          ,)  2*          b*  h+          +  ,          ,  -          .  
/         K  L         L  M          N  $O         &O  JP         LP  pQ         rQ  R         R  S         S  T         T  V         
V  .W         0W  TX         VX  zY         |Y  Z         Z  [         [  \         \  ^         ^  8_         :_  ^`         ``  a         a  b         b  c         c  d         d  f         f  Bg         Dg  hh         jh  i         i  j         j  k         k   m         m  &n         (n  Lo           >         @  d         f                        ֙         ؙ             "         $  H         J  n         p                                              ,         .  R         T  x         z             Ĩ         ƨ                        6         8  \         ^                        ΰ         а                        @         B  f                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               J!         R!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                      
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    <     -,-p     ,--     -G-    G-)5-    )5-C-    C-w-t    w-5-     5-5=-X    5=--    -b-p	    b-!-x
    !-/-
    /--    -ճ-\    ճ--<                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               M!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P              	          9  B          r  {                                              0  
          C  	          V	  0
          i
  C         |                                                                                 !  A         B  b         c                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        8"  Z#         [#  }$         ~$  %         %  &         &  '         '  )         )  1*         2*  U+         V+  y,         z,  -         -  .         .  /         /  	1         
1  -2         .2  Q3         R3  u4         v4  5         5  6         6  7         7  9         9  ):         *:  M;         N;  q<         r<  =         =  >         >  ?         ?  A         A  %B         &B  IC         JC  mD         nD  E         E  F         F  G         -l  Om         Pm  rn         sn  o         o  p         p  q         q  s         s  &t         't  Ju         Ku  nv         ov  w         w  x         x  y          y  z          z  {          #|  }          G}  >~          k~  b                      ց         ׁ                        B         C  f         g                        ҉         Ӊ           ړ             !         "  E         F  i         j                        ՛         ֛                        A         B  e         f                        ѣ         ң                        =         >  a         b                        ͫ         Ϋ                        9         :  ]         ^                        ɳ         ʳ                                           b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               J!         L!  1         1  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    h
         i
             :                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    h
         i
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     P
            0     -,-X     ,--     -A-    A-5-x    5-;D-|    ;D-w-\    w--    --h    -b-    b-n-	    n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               K!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         )  "          R  K          {  t                                              /  +	          `	  \
          
                                   $         %  E         F  f         \  |         }                                                                  b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            "         $  F         H  j         l                                              	          	  B
         D
  f         h                                                         >         @  b         d                                                         :         <  ^         `                                              !         !  6"          -  .          /  0          <0  21          b1  X2         .I  PJ         RJ  tK         vK  L         L  M         M  N         N  P         P  (Q         *Q  LR         NR  pS         rS  T         T  U         U  V         V   X         X  $Y         &Y  HZ         JZ  l[         n[  \         \  ]         ]  ^         ^  _         _  #a         $a  Ib         Jb  oc         pc  d         d  e         e  f         f  h         h  -i         .i  Sj         Tj  yk          k  {l         m  n         n  p         p  <q         =q  `r         ar  s         s  t         t  u         u  v         v  x         x  8y         9y  \z         ]z  {         {  |         |  }         }  ~         ~             4         5  X         Y  |         }             ą         Ņ                        0         1  T         U  x         y                                                      >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               K!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     4
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-;-x    ;-YJ-|    YJ-}-\    }--    --h    -b-    b-3n-	    3n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               D!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                       !  A         B  b         c                                                           =  6          f  _                                              :  D          s  }         y                                     !         !  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            $         &  J         L  p         r                                            
  .	         0	  T
         V
  z         |                                              8         :  ^         `                                                         B         D  h         j                                               &          (   L!         N!  r"         t"  #         #  $         $  %         %  '         '  1(         2(  W)         X)  }*         ~*  +         +  ,         ,  -         -  /         /  ;0         <0  a1         b1  2          ^  _          =_  /`          _`  Qa          a  sb          b  c         _j  k         k  l         l  m         m  n         n  p         p  Aq         Cq  gr         ir  s         s  t         t  u         u  v         w  %x         'x  Ky         My  qz         sz  {         {  |         |  }         }  	           /         1  U         W  {         }             Ǆ         Ʉ                        9         ;  _         a                        ь         ӌ                      	  .         /  T         U  z         {             ƕ         Ǖ                        8         9  ^         _                        Н         ѝ                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               E!         R!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     O
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -A-    A-5-x    5-;D-|    ;D-w-\    w--    --h    -b-    b-n-	    n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               K!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         )  #          R  L          {  u                                              /  *	          `	  [
          
                                   $         %  E         F  f         \  |         }                                                                  b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            #         $  G         H  k         l                                              	          	  C
         D
  g         h                                                         ?         @  c         d                                                         ;         <  _         `                                              !         !  7"         .I  QJ         RJ  uK         vK  L         L  M         M  N         N  P         P  )Q         *Q  MR         NR  qS         rS  T         T  U         U  V         V  X         X  %Y         &Y  IZ         JZ  m[         n[  \         \  ]         ]  ^         ^  _         _  "a         $a  Hb         Jb  nc         pc  d         d  e         e  f         f  h         h  ,i         .i  Rj         Tj  xk          l  m         m  n         n  p         p  ;q         =q  _r         ar  s         s  t         t  u         u  v         v  x         x  7y         9y  [z         ]z  {         {  |         |  }         }  ~         ~             3         5  W         Y  {         }             Å         Ņ                        /         1  S         U  w         y                                             	            /  %          U  K          {  q                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               G!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     3
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -;-    ;-;-x    ;-YJ-|    YJ-}-\    }--    --h    -b-    b-3n-	    3n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               F!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                       !  A         B  b         c                                                           =  7          f  `                                    
          :  C          s  |         y                                     !         !  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            %         &  K         L  q         r                                   	         
  /	         0	  U
         V
  {         |                                              9         :  _         `                                                         C         D  i         j                                              '          (   M!         N!  s"         t"  #         #  $         $  %         %  
'         '  0(         2(  V)         X)  |*         ~*  +         +  ,         ,  -         -  /         /  :0         <0  `1         b1  2          c  d          d  e          	f  f          +g  h          Mh  ?i         _j  k         k  l         l  m         m  n         n  p         p  Bq         Cq  hr         ir  s         s  t         t  u         u   w         w  &x         'x  Ly         My  rz         sz  {         {  |         |  }         }  
           0         1  V         W  |         }             Ȅ         Ʉ                        :         ;  `         a                        Ҍ         ӌ                      	  -         /  S         U  y         {             ŕ         Ǖ                        7         9  ]         _                        ϝ         ѝ                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               I!         V!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     N
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -G-    G-:-x    :-I-|    I-|-\    |-+-    +--h    -cb-    cb-x-	    x--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               M!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P           =         >  ^         _  	         	  
         
           ]  }         ~                                                                     (            ;  "          N  5          a  q                                     !          E!  U"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                      "  B         D  d         f                                              	         	  0
         2
  R         T  t         v                                                          @         B  b         d                                                         .         0  P         R  r         t                                     !         "  1#         3#  S$          <  =          =  >          >  @          /@  8A          eA  nB         sG  H         H  I         I  J         J  K         K  M         M  =N         ?N  _O         aO  P         P  Q         Q  R         R  S         S  	U         U  +V         -V  MW         OW  oX         qX  Y         Y  Z         Z  [         [  \         \  ^         ^  ;_         =_  ]`         _`  a         a  b         b  c         c  d         d  f         	f  )g         +g  Kh         Mh  mi         oi  j         j  k         k  l         $  D         F  f         h                        ̕         Ε                        2         4  T         V  v         x                        ܞ         ޞ                        "  B         D  d         f                        ʦ         ̦                        0         2  R         T  t                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               .!         :!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 o
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-5-x    5-C-|    C-v-\    v-I-    I--h    -{b-    {b--n-	    -n--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               @!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P              
          9  C          r  |                                              0            C  	          V	  1
          i
  D         |                                                                                 <         =  ]         ^  ~                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            &         (  N         P  v         x                                              >	         @	  f
         h
                                              .         0  V         X  ~                                                        F         H  n         p                                              6         8  ^          `   !         !  "         "  $         $  :%         <%  b&         d&  '         '  (         (  )         )  +         +  *,         ,,  R-         T-  z.         |.  /         /  0         0  1         1  3         3  B4         D4  j5         l5  6         6  7         7  8         8  
:         :  2;         4;  Z<         \<  =         =  >         >  ?         ?  @         @  "B         $B  JC         LC  rD         tD  E         E  F          G  )H          \H  lI          I  J          J  K          %L  5M         Bm  hn         jn  o         o  p         p  q         q  s         
s  0t         2t  Xu         Zu  v         v  w         w  x         x  y         y   {         "{  H|         J|  p}         r}  ~         ~                                   8         :  `         b                        ؇         ڇ              (         *  P         R  x         z             Ȏ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               <!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     O
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -;-    ;-:-x    :-I-|    I-|-\    |-+-    +--h    -ib-    ib-x-	    x--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               I!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P           =         >  ^         _  	         	  
         
           b                                                                                 -            @  &          S  9          f  u                                      !          J!  Y"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            !         "  C         D  e         f                                              	         	  1
         2
  S         T  u         v                                                          A         B  c         d                                                         /         0  Q         R  s         t                                     !         "  2#         3#  T$          B  C          C  D          E  F          =F  FG         sG  H         H  I         I  J         J  K         K  M         M  >N         ?N  `O         aO  P         P  Q         Q  R         R  S         S  
U         U  ,V         -V  NW         OW  pX         qX  Y         Y  Z         Z  [         [  \         \  ^         ^  <_         =_  ^`         _`  a         a  b         b  c         c  d         d  f         	f  *g         +g  Lh         Mh  ni         oi  j         j  k         k  l          l  m         $  E         F  g         h                        ͕         Ε                        3         4  U         V  w         x                        ݞ         ޞ              !         "  C         D  e         f                        ˦         ̦                        1         2  S         T  u                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               0!         >!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 o
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-5-x    5-C-|    C-w-\    w-I-    I--h    -b-    b--n-	    -n--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               B!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P              	          9  B          r  {                                              0  
          C  	          V	  0
          i
  C         |                                                                                 !  A         B  b         c                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            '         (  O         P  w         x                                              ?	         @	  g
         h
                                              /         0  W         X                                                          G         H  o         p                                              7         8  _          `   !         !  "         "  $         $  ;%         <%  c&         d&  '         '  (         (  )         )  +         +  +,         ,,  S-         T-  {.         |.  /         /  0         0  1         1  3         3  C4         D4  k5         l5  6         6  7         7  8         8  :         :  3;         4;  [<         \<  =         =  >         >  ?         ?  @         @  #B         $B  KC         LC  sD         tD  E         E  F          hM  xN          N  O          O  P          1Q  AR          tR  S         Bm  in         jn  o         o  p         p  q         q  	s         
s  1t         2t  Yu         Zu  v         v  w         w  x         x  y         y  !{         "{  I|         J|  q}         r}  ~         ~                                   9         :  a         b                        ه         ڇ             )         *  Q         R  y         z             Ɏ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               >!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     N
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-;-x    ;-MJ-|    MJ-}-\    }--    --h    -Eb-    Eb-3-	    3--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ?!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                       !  A         B  b         c                       r                                      -          V  f                                              
            3  ,         y                                     !         !  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            0         2  b         d                                   *         ,  \         ^  	         	  
         
             $         &  V         X                                               P         R                                              J         L  |         ~                                   D          F   v!         x!  "         "  #          #  $          #%  3&          ]&  m'          '  (          (  )         F  G         G  &I         (I  XJ         ZJ  K         K  L         L  M         M   O         "O  RP         TP  Q         Q  R         R  S         S  U         U  LV         NV  ~W         W  X         X  Y         Y  [         [  F\         H\  x]         z]  ^         ^  _         _  a         a  @b         Bb  rc         tc  d         d  e         e  g         
g  :h         <h  li         ni  j         j  k         k  m         m  5n         6n  go         ho  p         p  q         q  r         r  /t         0t  au         bu  v         v  w         w  x         x  )z         *z  [{         \{  |         |  }         }  ~         ~  #         $  U         V                                              O         P                                              I         J  {                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               %!         ,!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   Z      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-5;-x    5;-I-|    I-}-\    }-o-    o--h    -ub-    ub-x-	    x--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               >!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 /	         0	  P
         Q
  q         g                                                                    =  ?          n  p                                              *  #          S  L           |   u!          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            )         *  S         T  }         ~                                   %         &  O	         P	  y
         z
                                   !         "  K         L  u         v                                              G         H  q         r                                              C         D  m         n               !         !  "         $  <%         >%  f&         h&  '         '  (         (  )         )  +         +  8,         :,  b-         d-  .         .  /         /  0         0  
2         2  43         63  ^4         `4  5         5  6         6  7         7  9         9  0:         2:  Z;         \;  <         <  =         =  >         >  @         @  ,A         .A  VB         XB  C         C  D         D  E         E  F          G  (H         :H  bI         dI  J         J  K         K  L         L  
N         N  4O         6O  ^P         `P  Q         Q  R         R  S         S  U         U  0V         2V  ZW         \W  X         X  Y         Y  Z         Z  \         \  ,]         .]  V^         X^  _         _  `         `  a         a  b          c  (d         *d  Re         Te  |f         ~f  g         g  h          w  x          x  y          y  z          "{  |          J|  B}         ʎ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               7!         J!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     --    -;-x    ;-MJ-|    MJ-}-\    }--    --h    -Kb-    Kb-3-	    3--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ;!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                       !  A         B  b         c                       r                                      ,          V  e                                              
            3  -         y                                     !         !  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            1         2  c         d                                   +         ,  ]         ^  	         	  
         
             %         &  W         X                                               Q         R                                              K         L  }         ~                                   E          F   w!         x!  "         "  #          *  +          E+  U,          ,  -          -  .          .  0         F  G         G  'I         (I  YJ         ZJ  K         K  L         L  M         M  !O         "O  SP         TP  Q         Q  R         R  S         S  U         U  MV         NV  W         W  X         X  Y         Y  [         [  G\         H\  y]         z]  ^         ^  _         _  a         a  Ab         Bb  sc         tc  d         d  e         e  	g         
g  ;h         <h  mi         ni  j         j  k         k  m         m  4n         6n  fo         ho  p         p  q         q  r         r  .t         0t  `u         bu  v         v  w         w  x         x  (z         *z  Z{         \{  |         |  }         }  ~         ~  "         $  T         V                                              N         P                                              H         J  z                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               '!         0!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   Y      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-5;-x    5;-I-|    I-}-\    }-o-    o--h    -{b-    {b-x-	    x--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               <!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 /	         0	  P
         Q
  q         g                                                                    =  >          n  o                                              *  $          S  M           |   v!          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            (         *  R         T  |         ~                                   $         &  N	         P	  x
         z
                                             "  J         L  t         v                                              F         H  p         r                                              B         D  l         n               !         !  "         $  =%         >%  g&         h&  '         '  (         (  )         )  +         +  9,         :,  c-         d-  .         .  /         /  0         0  2         2  53         63  _4         `4  5         5  6         6  7         7  9         9  1:         2:  [;         \;  <         <  =         =  >         >  @         @  -A         .A  WB         XB  C         C  D         D  E         E  F          G  )H         :H  cI         dI  J         J  K         K  L         L  N         N  5O         6O  _P         `P  Q         Q  R         R  S         S  U         U  1V         2V  [W         \W  X         X  Y         Y  Z         Z  \         \  -]         .]  W^         X^  _         _  `         `  a         a  b          c  )d         *d  Se         Te  }f         ~f  g         g  h          r}  j~          ~                                      
         ʎ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ;!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-s-     s--    -:-x    :-I-|    I-|-\    |-
-    
--P    -cb-    cb-c-	    c-s-D
    s-3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               6!         @!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P           =         >  ^         _  	         	  
         
           ]  }         ~                                                                     (            ;  "          N  5          a  q                                     !          E!  U"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            .         0  ^         `                                               N         P  ~	         	  
         
                        >         @  n         p                                    .         0  ^         `                                               N         P  ~                                              >!         @!  n"         %  &         &  ((         *(  X)         Z)  *         *  +         +  ,         ,  .         .  H/         J/  x0         z0  1         1  2         2  4         
4  85         :5  h6         j6  7         7  8         8  9         9  (;         *;  X<         Z<  =         =  >         >  ?         ?  A         A  HB         JB  xC         zC  D         D  E         E  G         
G  8H          R  S          S  T          U  V          2V  2W          \W  \X         h   j         j  0k         2k  `l         bl  m         m  n         n  o         o   q         "q  Pr         Rr  s         s  t         t  u         u  w         w  @x         Bx  py         ry  z         z  {         {   }         }  0~         2~  `         b                                              L         N  {         }             ۈ         ݈             ;                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               .!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-5-x    5-C-|    C-v-\    v--    --h    -]b-    ]b-x-	    x--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               >!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P              
          9  C          r  |                                              0            C  	          V	  1
          i
  D         |                                                                                 <         =  ]         ^  ~                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                             "            P  K         &  '         '  (         (  
*         *  6+         7+  b,         c,  -         -  .         .  /         /  1         1  >2         ?2  j3         k3  4         4  5         5  6         6  8         8  F9         G9  r:         s:  ;         ;  <         <  =         =  !?         #?  M@         O@  yA         {A  B         B  C         C  D         D  )F         +F  UG         WG  H         H  I         I  J         J  L         L  1M         3M  ]N         _N  O         O  P         P  Q         Q  S         S  9T         ;T  eU         gU  V         V  W         W  X         X  Z         Z  A[         C[  m\         o\  ]         ]  ^         ^  _         _  a         a  Ib         Kb  uc         wc  d         d  e         e  f         f  %h         'h  Qi         Si  }j         j  k         k  l          n  *o         ,o  Vp         Xp  q         q  r         r  s         s  u         u  2v         4v  ^w         `w  x         x  y         y  z         z  |         |  :}         <}  f~         h~                                              B         D  n         p             Ƈ         ȇ                         J         L  v         x             Ύ         Ў             &                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               .!         <!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-m-     m--    -:-x    :-I-|    I-|-\    |-
-    
--P    -cb-    cb-c-	    c-s-D
    s-3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               3!         >!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P           =         >  ^         _  	         	  
         
           b                                                                                 -            @  &          S  9          f  u                                      !          J!  Y"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            /         0  _         `                                               O         P  	         	  
         
                        ?         @  o         p                                    /         0  _         `                                               O         P                                                ?!         @!  o"         %  &         &  )(         *(  Y)         Z)  *         *  +         +  ,         ,  .         .  I/         J/  y0         z0  1         1  2         2  	4         
4  95         :5  i6         j6  7         7  8         8  9         9  );         *;  Y<         Z<  =         =  >         >  ?         ?  A         A  IB         JB  yC         zC  D         D  E         E  	G         
G  9H          X  Y          Y  Z          Z  [          \  ]          .]  .^         h  j         j  1k         2k  al         bl  m         m  n         n  o         o  !q         "q  Qr         Rr  s         s  t         t  u         u  w         w  Ax         Bx  qy         ry  z         z  {         {  }         }  1~         2~  a         b                                              M         N  |         }             ܈         ݈             <                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               .!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     --    -5-x    5-C-|    C-v-\    v--    --h    -cb-    cb-x-	    x--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               <!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P              	          9  B          r  {                                              0  
          C  	          V	  0
          i
  C         |                                                                                 !  A         B  b         c                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         ~  y                                   !  "          6"  1#         &  '         '  (         (  	*         *  5+         7+  a,         c,  -         -  .         .  /         /  1         1  =2         ?2  i3         k3  4         4  5         5  6         6  8         8  E9         G9  q:         s:  ;         ;  <         <  =         =  "?         #?  N@         O@  zA         {A  B         B  C         C  D         D  *F         +F  VG         WG  H         H  I         I  J         J  L         L  2M         3M  ^N         _N  O         O  P         P  Q         Q  S         S  :T         ;T  fU         gU  V         V  W         W  X         X  Z         Z  B[         C[  n\         o\  ]         ]  ^         ^  _         _  a         a  Jb         Kb  vc         wc  d         d  e         e  f         f  &h         'h  Ri         Si  ~j         j  k         k  l          n  +o         ,o  Wp         Xp  q         q  r         r  s         s  u         u  3v         4v  _w         `w  x         x  y         y  z         z  |         |  ;}         <}  g~         h~                                              C         D  o         p             Ǉ         ȇ                         K         L  w         x             ώ         Ў             '                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               2!         @!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,-m-     m--    -;-    ;-GJ-    GJ-}-d    }--    --p    -{b-    {b--
    --
    --t    --׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               5!         >!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                       !  A         B  b         c                       r                                      -          V  f                                              
            3  ,         y                                     !         !  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         D  4          a  Q          ~  n                                 %  &         &  '         '  )         
)  6*         8*  d+         f+  ,         ,  -         -  .         .  0         0  J1         L1  x2         z2  3         3  4         4  6         6  07         27  ^8         `8  9         9  :         :  ;         ;  =         =  D>         F>  r?         t?  @         @  A         A  B         B  *D         ,D  XE         ZE  F         F  G         G  H         H  J         J  >K         @K  lL         nL  M         M  N         N  O         O  %Q         &Q  SR         TR  S         S  T         T  U         U  W         W  9X         :X  gY         hY  Z         Z  [         [  \         \  ^          ^  M_         N_  {`         |`  a         a  b         b  d         d  3e         4e  af         bf  g         g  h         h  i         i  k         k  Gl         Hl  um         vm  n         n  o         o  p          q  ,r         .r  Zs         \s  t         t  u         u  v         v  x         x  @y         By  nz         pz  {         {  |         |  }         }  &         (  T         V                        ރ                      :         <  h         j             ĉ         Ɖ                       "  N         P  |         ~                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ;!         R!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    h
         i
             ;      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     P
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-I-     I--    -/;-x    /;-I-|    I-}-\    }-,-    ,--h    -ib-    ib--	    --h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               !         0!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 /	         0	  P
         Q
  q         g                                                                    =  ?          n  p                                              *  #          S  L           |   u!          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            -         .  [         \                                              A         B  o	         p	  
         
                        '         (  U         V                                              ;         <  i         j                                   !         "  O         P  }         ~                        !         !  5"         6"  c#         %  &         &  '         '  %)         &)  S*         T*  +         +  ,         ,  -         -  /         /  90         :0  g1         h1  2         2  3         3  4         4  6          6  M7         N7  {8         |8  9         9  :         :  <         <  3=         4=  a>         b>  ?         ?  @         @  A         A  C         C  GD         HD  uE         vE  F         F  G         G  H         TI  J         J  K          K  L          L   N          0N  @O          pO  P          P  Q           ?         A  m         o             ɐ         ː             %         '  S         U                        ݗ         ߗ             9         ;  g         i             Ý         ŝ                      !  M         O  {         }             פ         ٤             3         5  a         c                                                      >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               5!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   #      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,-s-     s--    -;-    ;-GJ-    GJ-}-d    }-!-    !--p    -{b-    {b--
    --
    --t    --׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               7!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                       !  A         B  b         c                       r                                      ,          V  e                                              
            3  -         y                                     !         !  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                             ,             I   9!         %  &         &  '         '  	)         
)  7*         8*  e+         f+  ,         ,  -         -  .         .  0         0  K1         L1  y2         z2  3         3  4         4  6         6  17         27  _8         `8  9         9  :         :  ;         ;  =         =  E>         F>  s?         t?  @         @  A         A  B         B  +D         ,D  YE         ZE  F         F  G         G  H         H  J         J  ?K         @K  mL         nL  M         M  N         N  O         O  $Q         &Q  RR         TR  S         S  T         T  U         U  
W         W  8X         :X  fY         hY  Z         Z  [         [  \         \  ^          ^  L_         N_  z`         |`  a         a  b         b  d         d  2e         4e  `f         bf  g         g  h         h  i         i  k         k  Fl         Hl  tm         vm  n         n  o         o  p          q  -r         .r  [s         \s  t         t  u         u  v         v  x         x  Ay         By  oz         pz  {         {  |         |  }         }  '         (  U         V                        ߃                      ;         <  i         j             ŉ         Ɖ             !         "  O         P  }         ~                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ;!         R!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    h
         i
             :      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     Q
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-C-     C--    -/;-x    /;-I-|    I-}-\    }-,-    ,--h    -cb-    cb--	    --h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               !         .!  1         1  |=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 /	         0	  P
         Q
  q         g                                                                    =  >          n  o                                              *  $          S  M           |   v!          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ,         .  Z         \                                              @         B  n	         p	  
         
                        &         (  T         V                                              :         <  h         j                                             "  N         P  |         ~                        !         !  4"         6"  b#         %  &         &  '         '  $)         &)  R*         T*  +         +  ,         ,  -         -  
/         /  80         :0  f1         h1  2         2  3         3  4         4  6          6  L7         N7  z8         |8  9         9  :         :  <         <  2=         4=  `>         b>  ?         ?  @         @  A         A  C         C  FD         HD  tE         vE  F         F  G         G  H         TI  J         J  K          Q   S          0S  @T          pT  U          U  V          V   X           @         A  n         o             ʐ         ː             &         '  T         U                        ޗ         ߗ             :         ;  h         i             ĝ         ŝ                       !  N         O  |         }             ؤ         ٤             4         5  b         c                                                      >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               1!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   $      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,-m-     m--    -5-    5-#D-    #D-qw-l    qw-4-    4--x    -cb-	    cb-	-
    	--
    -m-|    m-ճ-    ճ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               3!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         )  "          R  K          {  t                                              /  +	          `	  \
          
                                   $         %  E         F  f         \  |         }                                                                  b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       >  O          |                          	         D%  n&         JL  uM         vM  N         N  O         O  P         P  %R         &R  QS         RS  }T         ~T  U         U  V         V  X         X  -Y         .Y  YZ         ZZ  [         [  \         \  ]         ]  	_         
_  5`         6`  aa         ba  b         b  c         c  d         d  f         f  =g         >g  ih         jh  i         i  j         j  k         k  m         m  En         Fn  qo         ro  p         p  q         q  r         r  !t         "t  Mu         Nu  yv         zv  w         w  x         x  y         y  ){         *{  U|         V|  }         }  ~         ~                        1         2  ]         ^                                              9         :  e         f                                              A         B  m         n             ē         Ɠ                        H         J  t         v             ̚         Κ             $         &  P         R  |         ~             ԡ         ֡              ,         .  X         Z                        ܨ         ި           
  4         6  `         b                                              <         >  h         j                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               4!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    h
         i
             ;      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    h
         i
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            0     -,-X     ,--     --    -;-x    ;-AJ-|    AJ-}-\    }-u-    u--h    -{b-    {b--	    --h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               >!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                       !  A         B  b         c                                                           =  6          f  _                                              :  D          s  }         y                                     !         !  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            .         0  ^         `                                               N         P  ~	         	  
         
                        >         @  n         p                                    .         0  ^         `                                               N         P  ~                                              >!         @!  n"         p"  #         D%  s&         t&  '         '  (         (  *         *  3+         4+  c,         d,  -         -  .         .  /         /  #1         $1  S2         T2  3         3  4         4  5         5  7         7  C8         D8  s9         t9  :         :  ;         ;  =         =  3>         4>  c?         d?  @         @  A         A  B         B  #D         $D  SE         TE  F         F  G         G  H          X  Y          +Z  >[          n[  \          \  ]          ]  _         ir  s         s  t         t  u         u  'w         )w  Wx         Yx  y         y  z         z  {         {  }         }  G~         I~  w         y             ׁ         ف           	  7         9  g         i             Ǉ         ɇ             '         )  W         Y                                              G         I  w         y             ה         ٔ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               D!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   #      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,-s-     s--    -5-    5-#D-    #D-qw-l    qw-4-    4--x    -cb-	    cb-	-
    	--
    -m-|    m-ճ-    ճ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               5!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         )  #          R  L          {  u                                              /  *	          `	  [
          
                                   $         %  E         F  f         \  |         }                                                                  b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         6  G          t              	          	            .  ?         D%  o&         JL  tM         vM  N         N  O         O  P         P  $R         &R  PS         RS  |T         ~T  U         U  V         V   X         X  ,Y         .Y  XZ         ZZ  [         [  \         \  ]         ]  _         
_  4`         6`  `a         ba  b         b  c         c  d         d  f         f  <g         >g  hh         jh  i         i  j         j  k         k  m         m  Dn         Fn  po         ro  p         p  q         q  r         r   t         "t  Lu         Nu  xv         zv  w         w  x         x  y         y  ({         *{  T|         V|  }         }  ~         ~                        0         2  \         ^                                              8         :  d         f                                              @         B  l         n             œ         Ɠ                        I         J  u         v             ͚         Κ             %         &  Q         R  }         ~             ա         ֡             -         .  Y         Z                        ݨ         ި  	         
  5         6  a         b                                              =         >  i         j                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               6!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    h
         i
             :      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    h
         i
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            0     -,-X     ,--     -)-    )-;-x    ;-AJ-|    AJ-}-\    }-u-    u--h    -{b-    {b--	    --h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               B!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                       !  A         B  b         c                                                           =  7          f  `                                    
          :  C          s  |         y                                     !         !  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            /         0  _         `                                               O         P  	         	  
         
                        ?         @  o         p                                    /         0  _         `                                               O         P                                                ?!         @!  o"         p"  #         D%  r&         t&  '         '  (         (  *         *  2+         4+  b,         d,  -         -  .         .  /         /  "1         $1  R2         T2  3         3  4         4  5         5  7         7  B8         D8  r9         t9  :         :  ;         ;  =         =  2>         4>  b?         d?  @         @  A         A  B         B  "D         $D  RE         TE  F         F  G         G  H          7_  J`          z`  a          a  b           c  d          Cd  Ve         ir  s         s  t         t  u         u  (w         )w  Xx         Yx  y         y  z         z  {         {  }         }  H~         I~  x         y             ؁         ف           	  8         9  h         i             ȇ         ɇ             (         )  X         Y                                              H         I  x         y             ؔ         ٔ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               B!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   $      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-g-     g--    -;-x    ;-AJ-|    AJ-}-\    }-96-    96--h    -ub-    ub-m-	    m--h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               0!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                       !  A         B  b         c                       r                                      -          V  f                                              
            3  ,         y                                     !         !  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            (         *  R         T  |         ~                                   $         &  N	         P	  x
         z
                                             "  J         L  t         v                                              F         H  p         r                                              B         D  l         n               !         !  "          x2  3          3  4          4  6          A6  T7          7  8         K  L         L  N         N  1O         2O  [P         \P  Q         Q  R         R  S         S  U         U  -V         .V  WW         XW  X         X  Y         Y  Z         Z  [          \  )]         *]  S^         T^  }_         ~_  `         `  a         a  b         b  %d         &d  Oe         Pe  yf         zf  g         g  h         h  i         i  !k         "k  Kl         Ll  um         vm  n         n  o           6         8  `         b                        ޚ                    
  2         4  \         ^                        ڡ         ܡ             .         0  X         Z                        ֨         ب              *         ,  T         V  ~                      ү         ԯ             &         (  P         R  z         |             ζ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               I!         J!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     4
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -/-    /-#;-    #;-I-    I-}-d    }--    -U-0    U--    -b-H	    b-c-P
    c-I-
    I-'׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               O!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 /	         0	  P
         Q
  q         g                                                                    =  ?          n  p                                              *  #          S  L           |   u!          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q   w  operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       \            1         2  c         d                                   +         ,  ]         ^  	         	  
         
             %         &  W         X                                               Q         R                                              K         L  }         ~                                   E          F   w!         x!  "         "  #         #  %         %  ?&         @&  q'         r'  (         (  )         )  +         +  9,         :,  k-         l-  .         .  /         /  1         1  32         42  e3         f3  4         4  5         5  6         6  -8         .8  _9         `9  :         :  ;         ;  <         <  '>         (>  Y?         Z?  @         @  A         A  B         B  !D         "D  SE         TE  F         F  G         G  H         H  J         ir  s         s  t         t  u         u  /w         1w  ax         cx  y         y  z         z  {         {  )}         +}  [~         ]~                                   #         %  U         W                                                  $          Q  V                                            I                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               k!         l!  !2         "2  >      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 s
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-g-     g--    -;-x    ;-AJ-|    AJ-}-\    }-96-    96--h    -b-    b-m-	    m--h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               2!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                       !  A         B  b         c                       r                                      ,          V  e                                              
            3  -         y                                     !         !  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            )         *  S         T  }         ~                                   %         &  O	         P	  y
         z
                                   !         "  K         L  u         v                                              G         H  q         r                                              C         D  m         n               !         !  "          8  9          
:  ;          M;  `<          <  =          =  >         K  L         L  N         N  0O         2O  ZP         \P  Q         Q  R         R  S         S  U         U  ,V         .V  VW         XW  X         X  Y         Y  Z         Z  [          \  (]         *]  R^         T^  |_         ~_  `         `  a         a  b         b  $d         &d  Ne         Pe  xf         zf  g         g  h         h  i         i   k         "k  Jl         Ll  tm         vm  n         n  o           7         8  a         b                        ߚ           	         
  3         4  ]         ^                        ۡ         ܡ             /         0  Y         Z                        ר         ب             +         ,  U         V                        ӯ         ԯ             '         (  Q         R  {         |             ϶                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               M!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     3
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -i -`     i -ְ-     ְ-1ܰ-    1ܰ-34-x    34-?-\    ?-Yh-<    Yh--    --    --    -b- 	    b-c-(
    c-I-
    I-'׳-
                   =         b      J      7   U   7   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               P!         R!  ]1                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 64 })  outputElements  5184    input0Elements  31104   input1Elements  24576   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     4                           #         b      J      7   M     operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       @                                            '	                      -         .  N         O  o          p  q                                              4  .          ]  W                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     4                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  1728    input0Elements  1728    invocations     1                           3         b      J      7   Q   w  operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       \            0         2  b         d                                   *         ,  \         ^  	         	  
         
             $         &  V         X                                               P         R                                              J         L  |         ~                                   D          F   v!         x!  "         "  #         #  %         %  >&         @&  p'         r'  (         (  )         )  +         +  8,         :,  j-         l-  .         .  /         /   1         1  22         42  d3         f3  4         4  5         5  6         6  ,8         .8  ^9         `9  :         :  ;         ;  <         <  &>         (>  X?         Z?  @         @  A         A  B         B   D         "D  RE         TE  F         F  G         G  H         H  J         ir  s         s  t         t  u         u  0w         1w  bx         cx  y         y  z         z  {         {  *}         +}  \~         ]~                                   $         %  V         W                                              P         Q                                                                                b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     4                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               g!         h!  2         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 s
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,-I-     I--    -5-    5-;D-    ;D-w-l    w-90-    90--x    -b-	    b--
    --
    -a-|    a-ճ-    ճ-!׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     f            !          R!  1          1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         '            N  C          u  j                                              )  -	          \	  `
          
                                   *         +  M         N  p         p                                                                            b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |        T$  v%         x%  &          ?  @          A  ,B          ^B  oC          C  D          D  E         vL  M         M  N         N  O         O  Q         Q  )R         *R  MS         NS  qT         rT  U         U  V         V  W         W  Y         Y  %Z         &Z  I[         J[  m\         n\  ]         ]  ^         ^  _         _  `         `  !b         "b  Ec         Fc  id         jd  e         e  f         f  g         g  h         h  j         j  Ak         Bk  el         fl  m         m  n         n  o         o  p         p  r         r  =s         >s  at         bt  u         u  v         v  w         w  x         x  z         z  9{         :{  ]|         ^|  }         }  ~         ~                                   5         6  Y         Z  }         ~             Ň         Ƈ                        1         2  U         V  y         z                                            
  ,         .  P         R  t         v                                              (         *  L         N  p         r                        ܠ         ޠ              $         &  H         J  l         n                        ب         ڨ                       "  D         F  h         j                        ԰                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               S!         T!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    j
         k
             @      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    j
         k
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,-7-     7--    -;-    ;-#J-    #J-q}-d    q}-a-    a--p    -{b-    {b--
    --
    --t    --                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     g            !          B!  1          1  z=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             "         #  E         F  h         i                                                           7  ,          ^  S                                              *  8          a  o         [  }         ~                          !         !  	#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         0  9          h  q                                             $  %         %  &         &  (         (  ')         ()  M*         N*  s+         t+  ,         ,  -         -  .         .  0         0  11         21  W2         X2  }3         ~3  4         4  5         5  6         6  8         8  ;9         <9  a:         b:  ;         ;  <         <  =         =  >         >  @          @  EA         FA  kB         lB  C         C  D         D  E         E  G         G  (H         *H  NI         PI  tJ         vJ  K         K  L         L  M         M  O         O  2P         4P  XQ         ZQ  ~R         R  S         S  T         T  U         U  W         W  <X         >X  bY         dY  Z         Z  [         [  \         \  ]         ]   _         "_  F`         H`  la         na  b         b  c         c  d         d  f         
f  0g         2g  Xh         Zh  i         i  j         j  k         k  l         l   n         "n  Ho         Jo  np         pp  q         q  r         r  s         s  u         u  ,v         .v  Rw         Tw  xx         zx  y         y  z         z  {         {  }         }  6~         8~  \         ^                        ΂         Ђ                        @         B  f         h                        ؊         ڊ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               M!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    j
         k
             @      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     +
                   
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            8     -,-h     ,-I-     I--    -5-    5-;D-    ;D-w-l    w-90-    90--x    -{b-	    {b--
    --
    -a-|    a-ճ-    ճ-!׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     e            !          R!  1          1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         '            N  D          u  k                                              )  ,	          \	  _
          
                                   *         +  M         N  p         p                                                                            b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |        T$  w%         x%  &          'F  8G          jG  {H          H  I          I  K          3K  DL         vL  M         M  N         N  O         O  Q         Q  (R         *R  LS         NS  pT         rT  U         U  V         V  W         W   Y         Y  $Z         &Z  H[         J[  l\         n\  ]         ]  ^         ^  _         _  `         `   b         "b  Dc         Fc  hd         jd  e         e  f         f  g         g  h         h  j         j  @k         Bk  dl         fl  m         m  n         n  o         o  p         p  r         r  <s         >s  `t         bt  u         u  v         v  w         w  x         x  z         z  8{         :{  \|         ^|  }         }  ~         ~                                   4         6  X         Z  |         ~             ć         Ƈ                        0         2  T         V  x         z                                   	         
  -         .  Q         R  u         v                                              )         *  M         N  q         r                        ݠ         ޠ             %         &  I         J  m         n                        ٨         ڨ             !         "  E         F  i         j                        հ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               O!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    j
         k
             @      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    j
         k
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,-=-     =--    -;-    ;-#J-    #J-q}-d    q}-a-    a--p    -{b-    {b--
    --
    --t    --                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     h            !          F!  1          1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             "         #  E         F  h         i                                                           7  -          ^  T                                               *  7          a  n         [  }         ~                          !         !  	#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         H  Q                                              (  1         $  %         %  &         &   (         (  &)         ()  L*         N*  r+         t+  ,         ,  -         -  .         .  
0         0  01         21  V2         X2  |3         ~3  4         4  5         5  6         6  8         8  :9         <9  `:         b:  ;         ;  <         <  =         =  >         >  @          @  DA         FA  jB         lB  C         C  D         D  E         E  G         G  )H         *H  OI         PI  uJ         vJ  K         K  L         L  M         M  O         O  3P         4P  YQ         ZQ  R         R  S         S  T         T  U         U  W         W  =X         >X  cY         dY  Z         Z  [         [  \         \  ]         ]  !_         "_  G`         H`  ma         na  b         b  c         c  d         d  	f         
f  1g         2g  Yh         Zh  i         i  j         j  k         k  l         l  !n         "n  Io         Jo  op         pp  q         q  r         r  s         s  u         u  -v         .v  Sw         Tw  yx         zx  y         y  z         z  {         {  }         }  7~         8~  ]         ^                        ς         Ђ                        A         B  g         h                        ي         ڊ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               M!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    j
         k
             @      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     ,
                   
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,-a-     a--    -:-x    :-cI-|    cI-|-\    |--    --h    -b-    b-qx-	    qx--h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     j            !          \!  1          
2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P           5         6  X         Y  {	         |	  
         
           X  z         {                                                                     -  	          @            S  /          f  j                                     !          B!  F"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            2         4  f         h                                   6         8  j         l  	         	  
         
             :         <  n         p                        
           >         @  r         t                                   B         D  v         x                                   F         H  z          |   !         !  "         "  $         $  J%         L%  ~&         &  '         '  (         (  *         *  N+         P+  ,         ,  -         -  .         .  0          0  R1         T1  2         2  3         3  4         4  "6         $6  V7         X7  8         8  9         9  :         :  &<         (<  Z=         \=  >         >  ?         ?  @         @  *B         ,B  ^C         `C  D         D  E         E  F         F  .H         ^H  I         I  J         J  K         K  -M         .M  aN         bN  O         O  P         P  Q         Q  1S         2S  eT         fT  U         U  V         V  X         X  5Y         6Y  iZ         jZ  [         [  \         \  ^         ^  9_         :_  m`         n`  a         a  b         b  	d         
d  =e         >e  qf         rf  g         g  h         h  j         j  Ak         Bk  ul          n  o          o  p          p  q          r  s          >s  +t                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               V!         Z!  2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 m
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            <     -,-p     ,-g-     g--    -4-    4-C-    C-v-t    v---     --C5-X    C5--    -3b-p	    3b-	-x
    	--
    -m-    m-ճ-\    ճ--<                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               5!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        7  E          n  |                                              &            9  	          L	  2
          _
  E         r                                                                                            B         C  e                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        V"  {#         }#  $         H  I         I  J         J  K         K  M         M  EN         GN  mO         oO  P         P  Q         Q  R         R  T         T  5U         7U  ]V         _V  W         W  X         X  Y         Y  Z         Z  %\         '\  M]         O]  u^         w^  _         _  `         `  a         a  c         c  =d         ?d  ee         ge  f         f  g         g  h         h  j         j  -k         /k  Ul         Wl  }m         m  n         n  o         o  p         p  r         r  Es         Gs  mt         ot  u         u  v         v  w         w  y         y  5z         7z  ]{         _{  |         |  }         }  ~         ~             %         '  M         O  u         w             Ņ         ǅ                        =         ?  e         g                        ݍ         ߍ             ,         .  S         U  {         }             ˔         ͔                        C         E  k         m                                              3         5  [         ]                                   գ  Τ                      %            M  F         u             ê         Ū                        ;                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     g            !          H!  1          1  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    j
         k
             @                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    l
         m
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            0     -,-X     ,-[-     [--    -:-x    :-]I-|    ]I-|-\    |-I-    I--h    -b-    b-kx-	    kx--h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     i            !          X!  1          2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P           5         6  X         Y  {	         |	  
         
           S  u         v                                                                     (            ;            N  +          a  d                                     	!          =!  @"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            3         4  g         h                                   7         8  k         l  	         	  
         
             ;         <  o         p                                   ?         @  s         t                                   C         D  w         x                                   G         H  {          |   !         !  "         "  $         $  K%         L%  &         &  '         '  (         (  *         *  O+         P+  ,         ,  -         -  .         .  0          0  S1         T1  2         2  3         3  4         4  #6         $6  W7         X7  8         8  9         9  :         :  '<         (<  [=         \=  >         >  ?         ?  @         @  +B         ,B  _C         `C  D         D  E         E  F         F  /H         ^H  I         I  J         J  K         K  ,M         .M  `N         bN  O         O  P         P  Q         Q  0S         2S  dT         fT  U         U  V         V   X         X  4Y         6Y  hZ         jZ  [         [  \         \  ^         ^  8_         :_  l`         n`  a         a  b         b  d         
d  <e         >e  pf         rf  g         g  h         h  j         j  @k         Bk  tl          bt  Ou          u  sv          v  w          w  x          x  y                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               T!         Z!  2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 m
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            <     -,-p     ,-m-     m--    -4-    4-C-    C-v-t    v-u--     u--5-X    5--    -3b-p	    3b-	-x
    	--
    -s-    s-ճ-\    ճ--<                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               7!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        7  D          n  {                                              &            9   	          L	  3
          _
  F         r                                                                                           =         >  `                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        V"  |#         }#  $         H  I         I  J         J  K         K  M         M  FN         GN  nO         oO  P         P  Q         Q  R         R  T         T  6U         7U  ^V         _V  W         W  X         X  Y         Y  Z         Z  &\         '\  N]         O]  v^         w^  _         _  `         `  a         a  c         c  >d         ?d  fe         ge  f         f  g         g  h         h  j         j  .k         /k  Vl         Wl  ~m         m  n         n  o         o  p         p  r         r  Fs         Gs  nt         ot  u         u  v         v  w         w  y         y  6z         7z  ^{         _{  |         |  }         }  ~         ~             &         '  N         O  v         w             ƅ         ǅ                        >         ?  f         g                        ލ         ߍ             -         .  T         U  |         }             ̔         ͔                        D         E  l         m                                              4         5  \         ]                        ԣ         գ             $         %  L         M  t          u  n                      Ū                                                        b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     h            !          H!  1          1  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    j
         k
             @                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    m
         n
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            0     -,-X     ,--     -5-    5-5-x    5-AD-|    AD-w-\    w--    --h    -Eb-    Eb-qx-	    qx--h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               G!         \!  1         
2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         '            N  C          u  j                                              )  -	          \	  `
          
                                   *         +  M         N  p         p                                                                            b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         @  .          `  N            n                                 $  %         %  '         '  8(         :(  j)         l)  *         *  +         +   -         -  2.         4.  d/         f/  0         0  1         1  2         2  ,4         .4  ^5         `5  6         6  7         7  8         8  &:         (:  X;         Z;  <         <  =         =  >         >   @         "@  RA         TA  B         B  C         C  D         D  F         F  LG         NG  ~H         H  I         I  K         K  BL         DL  tM         vM  N         N  O         O  
Q         Q  <R         >R  nS         pS  T         T  U         U  W         W  7X         8X  iY         jY  Z         Z  [         [  \          ]  1^         2^  c_         d_  `         `  a         a  b         b  +d         ,d  ]e         ^e  f         f  g         g  h         h  %j         &j  Wk         Xk  l         l  m         m  n         n  p          p  Qq         Rq  r         r  s         s  t         t  v         v  Kw         Lw  }x         ~x  y         y  z         z  |         |  E}         F}  w~         x~             ۀ         ܀             ?         @  q         r             Ն         ֆ             9         :  k         l             ό         Ќ             3         d                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     f            !          P!  1          1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 m
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-s-     s--    -;-x    ;-)J-|    )J-w}-\    w}-.-    .--h    -?b-    ?b-n-	    n--h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               6!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             "         #  E         F  h         i                                                           7  ,          ^  S                                              *  8          a  o         [  }         ~                          !         !  	#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            (         *  R         T  |         ~                                   $         &  N	         P	  x
         z
                                             "  J         L  t         v                                              F         H  p         r                                              B         D  l         n               !          #  $          $  %           &  '          0'  6(          `(  f)         @G  iH         jH  I         I  J         J  K         K  "M         $M  LN         NN  vO         xO  P         P  Q         Q  R         R  T          T  HU         JU  rV         tV  W         W  X         X  Y         Y  [         [  D\         F\  n]         p]  ^         ^  _         _  `         `  b         b  @c         Bc  jd         ld  e         e  f         f  g         g  i         i  <j         >j  fk         hk  l         l  m           ͏         Ώ             !         "  K         L  u         v             ɖ         ʖ                        G         H  q         r             ŝ         Ɲ                        C         D  m         n                      ¤                        ?         @  i         j                                              ;                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     g            !          N!  1          1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     O
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-5-x    5-AD-|    AD-w-\    w--    --h    -Eb-    Eb-qx-	    qx--h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               E!         X!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         '            N  D          u  k                                              )  ,	          \	  _
          
                                   *         +  M         N  p         p                                                                            b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                               @  .          `  N         $  %         %  '         '  9(         :(  k)         l)  *         *  +         +  -         -  3.         4.  e/         f/  0         0  1         1  2         2  -4         .4  _5         `5  6         6  7         7  8         8  ':         (:  Y;         Z;  <         <  =         =  >         >  !@         "@  SA         TA  B         B  C         C  D         D  F         F  MG         NG  H         H  I         I  K         K  CL         DL  uM         vM  N         N  O         O  Q         Q  =R         >R  oS         pS  T         T  U         U  W         W  6X         8X  hY         jY  Z         Z  [         [  \          ]  0^         2^  b_         d_  `         `  a         a  b         b  *d         ,d  \e         ^e  f         f  g         g  h         h  $j         &j  Vk         Xk  l         l  m         m  n         n  p          p  Pq         Rq  r         r  s         s  t         t  v         v  Jw         Lw  |x         ~x  y         y  z         z  |         |  D}         F}  v~         x~             ڀ         ܀             >         @  p         r             Ԇ         ֆ             8         :  j         l             Ό         Ќ              2         d                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     g            !          P!  1          1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 m
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-s-     s--    -;-x    ;-)J-|    )J-w}-\    w}-.-    .--h    -Kb-    Kb-n-	    n--h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               6!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             "         #  E         F  h         i                                                           7  -          ^  T                                               *  7          a  n         [  }         ~                          !         !  	#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            )         *  S         T  }         ~                                   %         &  O	         P	  y
         z
                                   !         "  K         L  u         v                                              G         H  q         r                                              C         D  m         n               !          )  *          *  +          +  ,           -  &.          P.  V/         @G  hH         jH  I         I  J         J  K         K  #M         $M  MN         NN  wO         xO  P         P  Q         Q  R         R  T          T  IU         JU  sV         tV  W         W  X         X  Y         Y  [         [  E\         F\  o]         p]  ^         ^  _         _  `         `  b         b  Ac         Bc  kd         ld  e         e  f         f  g         g  i         i  =j         >j  gk         hk  l         l  m           ̏         Ώ                       "  J         L  t         v             Ȗ         ʖ                        F         H  p         r             ĝ         Ɲ                        B         D  l         n                      ¤                        >         @  h         j                                              :                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     h            !          R!  1          1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     N
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-:-x    :-oI-|    oI-|-\    |-1-    1--h    --b-    -b--	    --h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ?!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P           5         6  X         Y  {	         |	  
         
           X  z         {                                                                     -  	          @            S  /          f  j                                     !          B!  F"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            .         0  ^         `                                               N         P  ~	         	  
         
                        >         @  n         p                                    .         0  ^         `                                               N         P  ~                                              >!         @!  n"         p"  #         #  $         $  %          &  .'         0'  ^(         `(  )         )  *         *  +         +  -          -  N.         P.  ~/         /  0         0  1         1  3         3  >4         @4  n5         p5  6         6  7         7  8          9  .:         0:  ^;         `;  <         <  =         =  >         >  @          @  NA         PA  ~B         B  C         C  D         D  F         F  >G         @G  nH         pH  I         I  J          ƅ  ҆          	            L  X                      Ҋ  ދ         d                      Ĕ             "         $  R         T                                              B         D  r         t             ҡ         ԡ             2         4  b         d             §         ħ             "         $  R         T                                              B         D  r                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     [                       D!  1          1  q=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   J      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-m-     m--    -4-x    4-C-|    C-v-\    v-	-    	--h    -cb-    cb-!n-	    !n--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               4!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        7  E          n  |                                              &            9  	          L	  2
          _
  E         r                                                                                            B         C  e                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            *         ,  V         X                                              2         4  ^	         `	  
         
                                   :         <  f         h                                              B         D  n         p                                               J         L  v         x                          !         !  &#          /  0          $1  :2          d2  z3          3  4          4  5         HI  rJ         tJ  K         K  L         L  M         M  "O         $O  NP         PP  zQ         |Q  R         R  S         S  T          U  *V         ,V  VW         XW  X         X  Y         Y  Z         Z  \         \  2]         4]  ^^         `^  _         _  `         `  a         a  c         c  :d         <d  fe         he  f         f  g         g  h         h  j         j  Bk         Dk  nl         pl  m         m  n         n  o         o  q          q  Jr         Lr  vs         xs  t         t  u         u  v         v  &x         (x  Ry         Ty  ~z         z  {         {  |         |  ~         ~  .         0  Z         \                        ރ           
           6         8  b         d                                              >         @  j         l                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               4!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     O
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-:-x    :-iI-|    iI-|-\    |-1-    1--h    --b-    -b--	    --h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               A!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P           5         6  X         Y  {	         |	  
         
           S  u         v                                                                     (            ;            N  +          a  d                                     	!          =!  @"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            /         0  _         `                                               O         P  	         	  
         
                        ?         @  o         p                                    /         0  _         `                                               O         P                                                ?!         @!  o"         p"  #         #  $         $  %          &  /'         0'  _(         `(  )         )  *         *  +         +  -          -  O.         P.  /         /  0         0  1         1  3         3  ?4         @4  o5         p5  6         6  7         7  8          9  /:         0:  _;         `;  <         <  =         =  >         >  @          @  OA         PA  B         B  C         C  D         D  F         F  ?G         @G  oH         pH  I         I  J            !          X  d                      ޏ            !  -         d             Ô         Ĕ             #         $  S         T                                              C         D  s         t             ӡ         ԡ             3         4  c         d             ç         ħ             #         $  S         T                                              C         D  s                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     \                       D!  1          1  r=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   I      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-m-     m--    -4-x    4-C-|    C-v-\    v-	-    	--h    -cb-    cb-n-	    n--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               4!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        7  D          n  {                                              &            9   	          L	  3
          _
  F         r                                                                                           =         >  `                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            +         ,  W         X                                              3         4  _	         `	  
         
                                   ;         <  g         h                                              C         D  o         p                                               K         L  w         x                          !         !  '#          $6  :7          d7  z8          8  9          9  :          $;  :<         HI  sJ         tJ  K         K  L         L  M         M  #O         $O  OP         PP  {Q         |Q  R         R  S         S  T          U  +V         ,V  WW         XW  X         X  Y         Y  Z         Z  \         \  3]         4]  _^         `^  _         _  `         `  a         a  c         c  ;d         <d  ge         he  f         f  g         g  h         h  j         j  Ck         Dk  ol         pl  m         m  n         n  o         o  q          q  Kr         Lr  ws         xs  t         t  u         u  v         v  'x         (x  Sy         Ty  z         z  {         {  |         |  ~         ~  /         0  [         \                        ߃                      7         8  c         d                                              ?         @  k         l                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               4!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     N
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-;-x    ;-5J-|    5J-}-\    }--    --h    -Eb-    Eb--	    --h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ?!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             "         #  E         F  h         i                       r  v                                              N  R            z                                              !           [  }         ~                          !         !  	#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ,         .  Z         \                                              @         B  n	         p	  
         
                        &         (  T         V                                              :         <  h         j                                             "  N         P  |         ~                        !         !  4"         6"  b#         d#  $         $  %         %  &         &  (         (  H)         J)  v*         x*  +         +  ,         ,   .         .  ./         0/  \0         ^0  1         1  2         2  3         3  5         5  B6         D6  p7         r7  8         8  9         9  :         :  (<         *<  V=         X=  >         >  ?         ?  @         @  B         B  <C         >C  jD         lD  E         E  F          HI  @J          tJ  lK          K  L          L  M          M  N         l  m         m  o          o  Mp         Np  {q         |q  r         r  s         s  u         u  3v         4v  aw         bw  x         x  y         y  z         z  |         |  G}         H}  u~         v~             р         Ҁ              -         .  [         \                                              A         B  o         p             ˍ         ̍             '                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               #!         2!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   J      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-y-     y--    -A;-x    A;-I-|    I-)}-\    )}-m-    m--h    -ib-    ib-x-	    x--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               8!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                               	  +	         ,	  N
         O
  q         q                                                                     S  M                                                          F  ;          m  b              !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            -         .  [         \                                              A         B  o	         p	  
         
                        '         (  U         V                                              ;         <  i         j                                   !         "  O         P  }         ~                        !         !  5"         6"  c#         d#  $         $  %         %  &         &  (         (  H)         J)  v*         x*  +         +  ,         ,   .         .  ./         0/  \0         ^0  1         1  2         2  3         3  5         5  B6         D6  p7         r7  8         8  9         9  :         :  (<         *<  V=         X=  >         >  ?         ?  @         @  B         B  <C         >C  jD         lD  E         E  F         F  G         G  "I         $I  PJ         RJ  ~K         K  L         L  M         M  O         
O  6P         8P  dQ         fQ  R         R  S         S  T         T  V         V  JW         LW  xX         zX  Y         Y  Z         Z  \         \  0]         2]  ^^         `^  _         _  `         `  a         a  c         c  Dd         Fd  re         te  f         f  g         g  h            ҄                      8  *          d  V                     ʐ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               5!         F!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-;-x    ;-5J-|    5J-}-\    }--    --h    -Kb-    Kb--	    --h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               =!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             "         #  E         F  h         i                       r  u                                              N  Q            {                                              !           [  }         ~                          !         !  	#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            -         .  [         \                                              A         B  o	         p	  
         
                        '         (  U         V                                              ;         <  i         j                                   !         "  O         P  }         ~                        !         !  5"         6"  c#         d#  $         $  %         %  &         &  (         (  I)         J)  w*         x*  +         +  ,         ,  .         .  //         0/  ]0         ^0  1         1  2         2  3         3  5         5  C6         D6  q7         r7  8         8  9         9  :         :  )<         *<  W=         X=  >         >  ?         ?  @         @  B         B  =C         >C  kD         lD  E         E  F          $O  P          PP  HQ          |Q  tR          R  S          S  T         l  m         m  o          o  Lp         Np  zq         |q  r         r  s         s  u         u  2v         4v  `w         bw  x         x  y         y  z         z  |         |  F}         H}  t~         v~             Ѐ         Ҁ              ,         .  Z         \                                              @         B  n         p             ʍ         ̍             &                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               '!         6!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   I      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-s-     s--    -A;-x    A;-I-|    I-)}-\    )}-m-    m--h    -ob-    ob-x-	    x--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               4!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                               	  +	         ,	  N
         O
  q         q                                                                     S  L                                                          F  <          m  c              !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ,         .  Z         \                                              @         B  n	         p	  
         
                        &         (  T         V                                              :         <  h         j                                             "  N         P  |         ~                        !         !  4"         6"  b#         d#  $         $  %         %  &         &  (         (  I)         J)  w*         x*  +         +  ,         ,  .         .  //         0/  ]0         ^0  1         1  2         2  3         3  5         5  C6         D6  q7         r7  8         8  9         9  :         :  )<         *<  W=         X=  >         >  ?         ?  @         @  B         B  =C         >C  kD         lD  E         E  F         F  G         G  #I         $I  QJ         RJ  K         K  L         L  M         M  	O         
O  7P         8P  eQ         fQ  R         R  S         S  T         T  V         V  KW         LW  yX         zX  Y         Y  Z         Z  \         \  1]         2]  _^         `^  _         _  `         `  a         a  c         c  Ed         Fd  se         te  f         f  g         g  h                        ڋ                      @  2          l  ^         ʐ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               7!         J!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-:-x    :-uI-|    uI-|-\    |--    --P    -{b-    {b-c-	    c-a-D
    a-3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               <!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P           5         6  X         Y  {	         |	  
         
           X  z         {                                                                     -  	          @            S  /          f  j                                     !          B!  F"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            *         ,  V         X                                              2         4  ^	         `	  
         
                                   :         <  f         h                                              B         D  n         p                                               J         L  v         x                          !         !  &#         (#  R$         T$  ~%         %  &         &  '         '  )         )  .*         0*  Z+         \+  ,         ,  -         -  .         .  
0         0  61         81  b2         d2  3         3  4         4  5         5  7         7  >8         @8  j9         l9  :         :  ;         ;  <         <  >         >  F?         H?  r@         t@  A         A  B         B  C         h  (j         *j  Tk         Vk  l         l  m         m  n         n  p         p  0q         2q  \r         ^r  s         s  t         t  u         u  w         w  8x         :x  dy         fy  z         z  {         {  |         |  ~         ~  @         B  l         n                      Ă                        C         E  o         q             ǉ         ɉ            Õ              ͗                                   ?  *                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               8!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 }
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     --    -4-x    4-C-|    C-v-\    v-i0-    i0--h    -Wb-    Wb-x-	    x--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               :!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        7  E          n  |                                              &            9  	          L	  2
          _
  E         r                                                                                            B         C  e                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                     @  3          `  S            s         %  	'         
'  9(         :(  i)         j)  *         *  +         +  ,         ,  ).         *.  Y/         Z/  0         0  1         1  2         2  4         4  I5         J5  y6         z6  7         7  8         8  	:         
:  9;         :;  i<         j<  =         =  >         >  ?         ?  )A         *A  YB         ZB  C         C  D         D  E         E  G         G  IH         JH  yI         h  i         i  *k         ,k  Zl         \l  m         m  n         n  o         o  q         q  Jr         Lr  zs         |s  t         t  u         u  
w         w  :x         <x  jy         ly  z         z  {         {  |         |  *~         ,~  Z         \                                              J         L  z         |             ڈ         ܈  
           :         <  j         P  ~                      ސ                      >         @  n         p             Ζ         Ж              .         0  ^         `                                               N         P  ~                      ޣ                      >         @  n         p             Ω         Щ              .         0  ^         `                                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               .!         8!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-:-x    :-oI-|    oI-|-\    |-= -    = --P    -{b-    {b-c-	    c-[-D
    [-3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ;!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P           5         6  X         Y  {	         |	  
         
           S  u         v                                                                     (            ;            N  +          a  d                                     	!          =!  @"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            +         ,  W         X                                              3         4  _	         `	  
         
                                   ;         <  g         h                                              C         D  o         p                                               K         L  w         x                          !         !  '#         (#  S$         T$  %         %  &         &  '         '  )         )  /*         0*  [+         \+  ,         ,  -         -  .         .  0         0  71         81  c2         d2  3         3  4         4  5         5  7         7  ?8         @8  k9         l9  :         :  ;         ;  <         <  >         >  G?         H?  s@         t@  A         A  B         B  C         h  )j         *j  Uk         Vk  l         l  m         m  n         n  p         p  1q         2q  ]r         ^r  s         s  t         t  u         u  w         w  9x         :x  ey         fy  z         z  {         {  |         |  ~         ~  A         B  m         n             Â         Ă                        D         E  p         q             ȉ         ɉ            ^  I          }  h                                  ڟ  Š                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               :!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 }
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-y-     y--    -4-x    4-C-|    C-v-\    v-i0-    i0--h    -]b-    ]b-x-	    x--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               6!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        7  D          n  {                                              &            9   	          L	  3
          _
  F         r                                                                                           =         >  `                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                              	  	           
           %  '         
'  8(         :(  h)         j)  *         *  +         +  ,         ,  (.         *.  X/         Z/  0         0  1         1  2         2  4         4  H5         J5  x6         z6  7         7  8         8  :         
:  8;         :;  h<         j<  =         =  >         >  ?         ?  (A         *A  XB         ZB  C         C  D         D  E         E  G         G  HH         JH  xI         h  i         i  +k         ,k  [l         \l  m         m  n         n  o         o  q         q  Kr         Lr  {s         |s  t         t  u         u  w         w  ;x         <x  ky         ly  z         z  {         {  |         |  +~         ,~  [         \                                              K         L  {         |             ۈ         ܈             ;         <  k         P                        ߐ                      ?         @  o         p             ϖ         Ж              /         0  _         `                                               O         P                        ߣ                      ?         @  o         p             ϩ         Щ              /         0  _         `                                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               0!         <!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     --    -;-    ;-;J-    ;J-}-l    }-G-    G--P    --    -b-h	    b--p
    -)-
    )--    -9׳-L                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               9!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             "         #  E         F  h         i                       r  v                                              N  R            z                                              !           [  }         ~                          !         !  	#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        %  1&         2&  ['         \'  (         (  )         )  *         *  ,         ,  --         .-  W.         X.  /         /  0         0  1         1  2          3  )4         *4  S5         T5  }6         ~6  7         7  8         8  9         9  %;         &;  O<         P<  y=         z=  >         >  ?         ?  @         @  !B         "B  KC         LC  uD         vD  E         E  F         F  G         G  I         I  FJ         HJ  pK         rK  L         L  M         M  N         N  P         P  BQ         DQ  lR         nR  S         S  T         T  U         U  W         W  >X         @X  hY         jY  Z         Z  [         [  \         \  ^         ^  :_         <_  d`         f`  a         a  b         b  c         c  e         e  6f         8f  `g         bg  h         h  i         i  j         j  l         
l  2m         4m  \n         ^n  o         o  p         p  q         q  s         s  .t         0t  Xu         Zu  v         v  w         w  x         x   z         z  *{         ,{  T|         V|  ~}         }  ~         ~                        &          (  #          R  M          |  w                      І  ˇ           "         $  L         N  v         x             ʍ                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               E!         ^!  1         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    j
         k
             @                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     H
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-a-     a--    -G;-x    G;-I-|    I-/}-\    /}-C-    C--h    -ib-    ib--	    --h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               )!         4!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                               	  +	         ,	  N
         O
  q         q                                                                     S  M                                                          F  ;          m  b              !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            1         2  c         d                                   +         ,  ]         ^  	         	  
         
             %         &  W         X                                               Q         R                                              K         L  }         ~                                   E          F   w!         x!  "         "  #          `1  j2          2  3          3  4          5  6          P6  Z7         G  H         H  I         I  -K         .K  _L         `L  M         M  N         N  O         O  'Q         (Q  YR         ZR  S         S  T         T  U         U  !W         "W  SX         TX  Y         Y  Z         Z  [         [  ]         ]  M^         N^  _         _  `         `  a         a  c         c  Gd         Hd  ye         ze  f         f  g         g  i         i  Aj         Bj  sk         tk  l         l  m         Pp  q         q  r         r  s         s  u         u  Hv         Jv  zw         |w  x         x  y         y  {         {  B|         D|  t}         v}  ~         ~             
           <         >  n         p             ҅         ԅ             6         8  h         j             ̋         ΋              0         2  b         d             Ƒ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               1!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     --    -;-    ;-;J-    ;J-}-l    }--    --P    --    -b-h	    b--p
    -)-
    )--    -9׳-L                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               9!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             "         #  E         F  h         i                       r  u                                              N  Q            {                                              !           [  }         ~                          !         !  	#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        %  0&         2&  Z'         \'  (         (  )         )  *         *  ,         ,  ,-         .-  V.         X.  /         /  0         0  1         1  2          3  (4         *4  R5         T5  |6         ~6  7         7  8         8  9         9  $;         &;  N<         P<  x=         z=  >         >  ?         ?  @         @   B         "B  JC         LC  tD         vD  E         E  F         F  G         G  I         I  GJ         HJ  qK         rK  L         L  M         M  N         N  P         P  CQ         DQ  mR         nR  S         S  T         T  U         U  W         W  ?X         @X  iY         jY  Z         Z  [         [  \         \  ^         ^  ;_         <_  e`         f`  a         a  b         b  c         c  e         e  7f         8f  ag         bg  h         h  i         i  j         j  	l         
l  3m         4m  ]n         ^n  o         o  p         p  q         q  s         s  /t         0t  Yu         Zu  v         v  w         w  x         x  z         z  +{         ,{  U|         V|  }         }  ~         ~                        '         (  Q         R  {         |             φ         І                        $            N  I          x  s                                          b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               E!         ^!  1         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    j
         k
             @                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     I
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-[-     [--    -G;-x    G;-I-|    I-/}-\    /}-C-    C--h    -cb-    cb--	    --h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               &!         2!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                               	  +	         ,	  N
         O
  q         q                                                                     S  L                                                          F  <          m  c              !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            0         2  b         d                                   *         ,  \         ^  	         	  
         
             $         &  V         X                                               P         R                                              J         L  |         ~                                   D          F   v!         x!  "         "  #          7  8          8  9          :  ;          @;  J<          |<  =         G  H         H  I         I  ,K         .K  ^L         `L  M         M  N         N  O         O  &Q         (Q  XR         ZR  S         S  T         T  U         U   W         "W  RX         TX  Y         Y  Z         Z  [         [  ]         ]  L^         N^  ~_         _  `         `  a         a  c         c  Fd         Hd  xe         ze  f         f  g         g  i         i  @j         Bj  rk         tk  l         l  m         Pp  q         q  r         r  s         s  u         u  Iv         Jv  {w         |w  x         x  y         y  {         {  C|         D|  u}         v}  ~         ~                        =         >  o         p             Ӆ         ԅ             7         8  i         j             ͋         ΋              1         2  c         d             Ǒ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               -!         @!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     -)-    )-5-    5-wD-    wD-w-l    w-72-    72--x    -b-	    b-!-
    !-/-
    /--|    -ճ-    ճ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ;!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         '            N  C          u  j                                              )  *	          \	  ]
          
                                   *         +  M         N  p         p                                              A                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |        $  %          %  '          7'  H(          z(  )          )  *           +  ,         K  L         L  N         N  EO         FO  mP         nP  Q         Q  R         R  S         S  U         U  5V         6V  ]W         ^W  X         X  Y         Y  Z         Z  [         [  %]         &]  M^         N^  u_         v_  `         `  a         a  b         b  d         d  =e         >e  ef         ff  g         g  h         h  i         i  k         k  -l         .l  Um         Vm  }n         ~n  o         o  p         p  q         q  s         s  Et         Ft  mu         nu  v         v  w         w  x         x  z         z  5{         6{  ]|         ^|  }         }  ~         ~                        %         &  M         N  u         v             ņ         Ɔ                        =         >  e         f                        ݎ         ގ             -         .  T         V  |         ~             ̕         Ε                        D         F  l         n                                              4         6  \         ^                        Ԥ         ֤             $         &  L         N  t         v             ī         ƫ                        <         >  d         f                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               <!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    j
         k
             @      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    j
         k
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     \
            0     -,-X     ,--     -/-    /-;-x    ;-J-|    J-}-\    }--    --h    -{b-    {b--	    --h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               B!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             "         #  E         F  h         i                                                           7  ,          ^  S                                              *  3          a  j         ~                                     %"         &"  O#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            2         4  f         h                                   6         8  j         l  	         	  
         
             :         <  n         p                        
           >         @  r         t                                   B         D  v         x                                   F         H  z          |   !         !  "         "  $         $  %          &  3'         4'  g(         h(  )         )  *         *  ,         ,  7-         8-  k.         l.  /         /  0         0  2         2  ;3         <3  o4         p4  5         5  6         6  8         8  ?9         @9  s:         t:  ;         ;  <         <  >         >  C?         D?  w@         x@  A         A  B         B  D         D  GE         HE  {F         |F  G         G  H          
e  f          Mf  ^g          g  h          h  i          j  'k         "o  Tp         Vp  q         q  r         r  s         s  $u         &u  Xv         Zv  w         w  x         x  y         y  ({         *{  \|         ^|  }         }  ~         ~             ,         .  `         b             Ȅ         ʄ             0         2  d         f             ̊         Ί              4         6  h         j             А         Ґ             8                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               6!         J!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     -)-    )-5-    5-qD-    qD-w-l    w-72-    72--x    -{b-	    {b-!-
    !-/-
    /--|    -ճ-    ճ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ;!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         '            N  D          u  k                                              )  +	          \	  ^
          
                                   *         +  M         N  p         p                                              <                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |        $  %          C,  T-          -  .          .  /          0  1          O1  `2         K  L         L  N         N  DO         FO  lP         nP  Q         Q  R         R  S         S  U         U  4V         6V  \W         ^W  X         X  Y         Y  Z         Z  [         [  $]         &]  L^         N^  t_         v_  `         `  a         a  b         b  d         d  <e         >e  df         ff  g         g  h         h  i         i  k         k  ,l         .l  Tm         Vm  |n         ~n  o         o  p         p  q         q  s         s  Dt         Ft  lu         nu  v         v  w         w  x         x  z         z  4{         6{  \|         ^|  }         }  ~         ~                        $         &  L         N  t         v             Ć         Ɔ                        <         >  d         f                        ܎         ގ             ,         .  U         V  }         ~             ͕         Ε                        E         F  m         n                                              5         6  ]         ^                        դ         ֤             %         &  M         N  u         v             ū         ƫ                        =         >  e         f                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               :!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    j
         k
             @      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    j
         k
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     [
            0     -,-X     ,--     -5-    5-;-x    ;-J-|    J-}-\    }--    --h    -ub-    ub--	    --h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               D!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             "         #  E         F  h         i                                                           7  -          ^  T                                              *  2          a  i         y                                     $"         %"  O#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            3         4  g         h                                   7         8  k         l  	         	  
         
             ;         <  o         p                                   ?         @  s         t                                   C         D  w         x                                   G         H  {          |   !         !  "         "  $         $  %          &  2'         4'  f(         h(  )         )  *         *  ,         ,  6-         8-  j.         l.  /         /  0         0  2         2  :3         <3  n4         p4  5         5  6         6  
8         8  >9         @9  r:         t:  ;         ;  <         <  >         >  B?         D?  v@         x@  A         A  B         B  D         D  FE         HE  zF         |F  G         G  H          Yk  jl          l  m          m  n         "o  Up         Vp  q         q  r         r  s         s  %u         &u  Yv         Zv  w         w  x         x  y         y  ){         *{  ]|         ^|  }         }  ~         ~             -         .  a         b             Ʉ         ʄ             1         2  e         f             ͊         Ί             5         6  i         j             ѐ         Ґ             9            ʔ                               >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               4!         J!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -M-    M-;-x    ;-GJ-|    GJ-}-\    }-4-    4--h    -b-    b-	n-	    	n--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               O!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             "         #  E         F  h         i                       r  k                                              N  G            z                                              !           [  }         ~                          !         !  	#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            $         &  J         L  p         r                                            
  .	         0	  T
         V
  z         |                                              8         :  ^         `                                                         B         D  h         j                                               &          (   L!         N!  r"          :/  >0          p0  t1          1  2          2  3          4  5         K  L         L  M          N  %O         &O  KP         LP  qQ         rQ  R         R  S         S  T         T  	V         
V  /W         0W  UX         VX  {Y         |Y  Z         Z  [         [  \         \  ^         ^  9_         :_  _`         ``  a         a  b         b  c         c  d         d  f         f  Cg         Dg  ih         jh  i         i  j         j  k         k  m         m  'n         (n  Mo           ?         @  e         f                        י         ؙ             #         $  I         J  o         p                                              -         .  S         T  y         z             Ũ         ƨ                        7         8  ]         ^                        ϰ         а                        A         B  g                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               P!         V!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                      
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    <     -,-p     ,--     -G-    G-5-    5-C-    C-v-t    v-5-     5-5=-X    5=--    -b-p	    b-'-x
    '-5-
    5--    -ճ-\    ճ--<                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               M!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        7  =          n  t                                              &            9  	          L	  *
          _
  =         r                                                                                            B         C  e                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h         8"  ,#          [#  O$         ~$  %         %  &         &  '         '  )         )  0*         2*  T+         V+  x,         z,  -         -  .         .  /         /  1         
1  ,2         .2  P3         R3  t4         v4  5         5  6         6  7         7  9         9  (:         *:  L;         N;  p<         r<  =         =  >         >  ?         ?   A         A  $B         &B  HC         JC  lD         nD  E         E  F          F  G          -l  !m          Pm  Dn         sn  o         o  p         p  q         q  s         s  %t         't  Iu         Ku  mv         ov  w         w  x         x  y         y  z         z  !|         #|  E}         G}  i~         k~                        Ձ         ׁ                        A         C  e         g                        щ         Ӊ           ړ                       "  D         F  h         j                        ԛ         ֛                        @         B  d         f                        У         ң                        <         >  `         b                        ̫         Ϋ                        8         :  \         ^                        ȳ         ʳ                                           b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               L!         P!  1         1  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    j
         k
             @                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    j
         k
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     F
            0     -,-X     ,--     -M-    M-;-x    ;-GJ-|    GJ-}-\    }-4-    4--h    -b-    b-	n-	    	n--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               O!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             "         #  E         F  h         i                       r  j                                              N  F            {                                              !           [  }         ~                          !         !  	#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            %         &  K         L  q         r                                   	         
  /	         0	  U
         V
  {         |                                              9         :  _         `                                                         C         D  i         j                                              '          (   M!         N!  s"          H5  L6          ~6  7          7  8          8  9           :  $;         K  L         L  M          N  $O         &O  JP         LP  pQ         rQ  R         R  S         S  T         T  V         
V  .W         0W  TX         VX  zY         |Y  Z         Z  [         [  \         \  ^         ^  8_         :_  ^`         ``  a         a  b         b  c         c  d         d  f         f  Bg         Dg  hh         jh  i         i  j         j  k         k   m         m  &n         (n  Lo           >         @  d         f                        ֙         ؙ             "         $  H         J  n         p                                              ,         .  R         T  x         z             Ĩ         ƨ                        6         8  \         ^                        ΰ         а                        @         B  f                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               N!         V!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                      
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    <     -,-p     ,--     -M-    M-5-    5-C-    C-v-t    v-5-     5-5=-X    5=--    -b-p	    b-'-x
    '-5-
    5--    -ճ-\    ճ--<                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               Q!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        7  <          n  s                                              &            9  	          L	  +
          _
  >         r                                                                                           =         >  `                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        8"  Z#         [#  }$         ~$  %         %  &         &  '         '  )         )  1*         2*  U+         V+  y,         z,  -         -  .         .  /         /  	1         
1  -2         .2  Q3         R3  u4         v4  5         5  6         6  7         7  9         9  ):         *:  M;         N;  q<         r<  =         =  >         >  ?         ?  A         A  %B         &B  IC         JC  mD         nD  E         E  F         F  G         -l  Om         Pm  rn          sn  ho          o  p          p  q          q  r          s  s         't  Ju         Ku  nv         ov  w         w  x         x  y         y  z         z  "|         #|  F}         G}  j~         k~                        ց         ׁ                        B         C  f         g                        ҉         Ӊ           ړ             !         "  E         F  i         j                        ՛         ֛                        A         B  e         f                        ѣ         ң                        =         >  a         b                        ͫ         Ϋ                        9         :  ]         ^                        ɳ         ʳ                                           b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               N!         P!  1         1  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    j
         k
             @                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    j
         k
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     E
            0     -,-X     ,--     -G-    G-5-x    5-YD-|    YD-w-\    w--    --h    -b-    b-n-	    n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               O!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         '            N  C          u  j                                              )  %	          \	  X
          
                                   *         +  M         N  p         p                                                                            b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            "         $  F         H  j         l                                              	          	  B
         D
  f         h                                                         >         @  b         d                                                         :         <  ^         `                                              !         !  6"         .I  PJ         RJ  tK         vK  L         L  M         M  N         N  P         P  (Q         *Q  LR         NR  pS         rS  T         T  U         U  V         V   X         X  $Y         &Y  HZ         JZ  l[         n[  \         \  ]         ]  ^         ^  _         _  #a         $a  Ib         Jb  oc         pc  d         d  e         e  f         f  h         h  -i         .i  Sj         Tj  yk         m  n         n  p         p  <q         =q  `r         ar  s         s  t         t  u         u  v         v  x         x  8y         9y  \z         ]z  {         {  |         |  }         }  ~         ~             4         5  X         Y  |         }             ą         Ņ                        0         1  T         U  x         y                                                         Ǖ                                    9  -                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               O!         T!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     0
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -A-    A-;-x    ;-AJ-|    AJ-}-\    }--    --h    -b-    b-3n-	    3n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               H!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             "         #  E         F  h         i                                                           7  ,          ^  S                                              *  0          a  g         [  }         ~                          !         !  	#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            $         &  J         L  p         r                                            
  .	         0	  T
         V
  z         |                                              8         :  ^         `                                                         B         D  h         j                                               &          (   L!         N!  r"         t"  #         #  $         $  %         %  '         '  1(         2(  W)         X)  }*         ~*  +         +  ,         ,  -         -  /         /  ;0         <0  a1         b1  2          R  S          S  T          U  U          -V  W          OW  ?X         _j  k         k  l         l  m         m  n         n  p         p  Aq         Cq  gr         ir  s         s  t         t  u         u  v         w  %x         'x  Ky         My  qz         sz  {         {  |         |  }         }  	           /         1  U         W  {         }             Ǆ         Ʉ                        9         ;  _         a                        ь         ӌ                      	  .         /  T         U  z         {             ƕ         Ǖ                        8         9  ^         _                        Н         ѝ                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               I!         V!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     O
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -G-    G-5-x    5-YD-|    YD-w-\    w-q-    q--h    -b-    b-n-	    n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               O!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         '            N  D          u  k                                              )  $	          \	  W
          
                                   *         +  M         N  p         p                                                                            b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            #         $  G         H  k         l                                              	          	  C
         D
  g         h                                                         ?         @  c         d                                                         ;         <  _         `                                              !         !  7"         .I  QJ         RJ  uK         vK  L         L  M         M  N         N  P         P  )Q         *Q  MR         NR  qS         rS  T         T  U         U  V         V  X         X  %Y         &Y  IZ         JZ  m[         n[  \         \  ]         ]  ^         ^  _         _  "a         $a  Hb         Jb  nc         pc  d         d  e         e  f         f  h         h  ,i         .i  Rj         Tj  xk         m  n         n  p         p  ;q         =q  _r         ar  s         s  t         t  u         u  v         v  x         x  7y         9y  [z         ]z  {         {  |         |  }         }  ~         ~             3         5  W         Y  {         }             Å         Ņ                        /         1  S         U  w         y                                             _  S            y                      ѝ  Ş                               >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               K!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     /
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -A-    A-;-x    ;-AJ-|    AJ-}-\    }--    --h    -b-    b-3n-	    3n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               J!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             "         #  E         F  h         i                                                           7  -          ^  T                                              *  /          a  f         [  }         ~                          !         !  	#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            %         &  K         L  q         r                                   	         
  /	         0	  U
         V
  {         |                                              9         :  _         `                                                         C         D  i         j                                              '          (   M!         N!  s"         t"  #         #  $         $  %         %  
'         '  0(         2(  V)         X)  |*         ~*  +         +  ,         ,  -         -  /         /  :0         <0  `1         b1  2          qX  aY          Y  Z          Z  [          [  \          \  ]         _j  k         k  l         l  m         m  n         n  p         p  Bq         Cq  hr         ir  s         s  t         t  u         u   w         w  &x         'x  Ly         My  rz         sz  {         {  |         |  }         }  
           0         1  V         W  |         }             Ȅ         Ʉ                        :         ;  `         a                        Ҍ         ӌ                      	  -         /  S         U  y         {             ŕ         Ǖ                        7         9  ]         _                        ϝ         ѝ                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               M!         Z!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     N
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -M-    M-:-x    :-I-|    I-|-\    |-;-    ;--h    -Wb-    Wb-3-	    3--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               Q!         X!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P           5         6  X         Y  {	         |	  
         
           X  z         {                                                                     -            @  $          S  7          f  r                                     !          B!  N"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            2         4  f         h                                   6         8  j         l  	         	  
         
             :         <  n         p                        
           >         @  r         t                                   B         D  v         x                                   F         H  z          |   !         !  "         "  $         $  J%         L%  ~&         &  '         '  (         (  *         *  N+         P+  ,         ,  -         -  .         .  0          0  R1         T1  2         2  3         3  4         4  "6         $6  V7         X7  8         8  9         9  :         :  &<         (<  Z=         \=  >         >  ?         ?  @         @  *B         ,B  ^C         `C  D         D  E         E  F         F  .H         0H  bI         dI  J         J  K         K  L          M  2N         4N  fO         hO  P         P  Q         Q  S         S  6T         8T  jU         lU  V         W  QX         SX  Y         Y  Z         Z  [         [  !]         #]  U^         W^  _         _  `         `  a         a  %c         'c  Yd         [d  e         e  f         f  g         g  )i         +i  ]j          zk  l          l  m           n  o          Co  Wp          p  q         T                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               *!         6!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   R      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-4-x    4-C-|    C-v-\    v-I-    I--h    -b-    b--n-	    -n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               D!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        7  =          n  t                                              &            9  	          L	  *
          _
  =         r                                                                                            B         C  e                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            &         (  N         P  v         x                                              >	         @	  f
         h
                                              .         0  V         X  ~                                                        F         H  n         p                                              6         8  ^          `   !         !  "         "  $         $  :%         <%  b&         d&  '         '  (         (  )         )  +         +  *,         ,,  R-         T-  z.         |.  /         /  0         0  1         1  3         3  B4         D4  j5         l5  6         6  7         7  8         8  
:         :  2;         4;  Z<         \<  =         =  >         >  ?         ?  @         @  "B         $B  JC         LC  rD         tD  E         E  F          ^  _          _  `          a  b          Bb  Bc          tc  td         Bm  hn         jn  o         o  p         p  q         q  s         
s  0t         2t  Xu         Zu  v         v  w         w  x         x  y         y   {         "{  H|         J|  p}         r}  ~         ~                                   8         :  `         b                        ؇         ڇ              (         *  P         R  x         z             Ȏ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               @!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     O
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -G-    G-:-x    :-I-|    I-|-\    |-;-    ;--h    -]b-    ]b-3-	    3--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               M!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P           5         6  X         Y  {	         |	  
         
           S  u         v                                                                     (            ;             N  3          a  l                                     !          =!  H"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            3         4  g         h                                   7         8  k         l  	         	  
         
             ;         <  o         p                                   ?         @  s         t                                   C         D  w         x                                   G         H  {          |   !         !  "         "  $         $  K%         L%  &         &  '         '  (         (  *         *  O+         P+  ,         ,  -         -  .         .  0          0  S1         T1  2         2  3         3  4         4  #6         $6  W7         X7  8         8  9         9  :         :  '<         (<  [=         \=  >         >  ?         ?  @         @  +B         ,B  _C         `C  D         D  E         E  F         F  /H         0H  cI         dI  J         J  K         K  L          M  3N         4N  gO         hO  P         P  Q         Q  S         S  7T         8T  kU         lU  V         W  RX         SX  Y         Y  Z         Z  [         [  "]         #]  V^         W^  _         _  `         `  a         a  &c         'c  Zd         [d  e         e  f         f  g         g  *i         +i  ^j          q  r          s   t          Ot  cu          u  v          v  w         T                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ,!         :!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   Q      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-4-x    4-C-|    C-v-\    v-I-    I--h    -b-    b--n-	    -n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               F!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        7  <          n  s                                              &            9  	          L	  +
          _
  >         r                                                                                           =         >  `                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            '         (  O         P  w         x                                              ?	         @	  g
         h
                                              /         0  W         X                                                          G         H  o         p                                              7         8  _          `   !         !  "         "  $         $  ;%         <%  c&         d&  '         '  (         (  )         )  +         +  +,         ,,  S-         T-  {.         |.  /         /  0         0  1         1  3         3  C4         D4  k5         l5  6         6  7         7  8         8  :         :  3;         4;  [<         \<  =         =  >         >  ?         ?  @         @  #B         $B  KC         LC  sD         tD  E         E  F          d  e          e  f          
g  
h          <h  <i          ni  nj         Bm  in         jn  o         o  p         p  q         q  	s         
s  1t         2t  Yu         Zu  v         v  w         w  x         x  y         y  !{         "{  I|         J|  q}         r}  ~         ~                                   9         :  a         b                        ه         ڇ             )         *  Q         R  y         z             Ɏ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               B!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     N
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-;-x    ;-;J-|    ;J-}-\    }--    --h    -Eb-    Eb---	    ---h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               C!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             "         #  E         F  h         i                       r  ~                                    #          N  Z            z                                              !           [  }         ~                          !         !  	#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            0         2  b         d                                   *         ,  \         ^  	         	  
         
             $         &  V         X                                               P         R                                              J         L  |         ~                                   D          F   v!         x!  "         "  #          -0  ;1          g1  u2          2  3          3  4          5  #6         F  G         G  &I         (I  XJ         ZJ  K         K  L         L  M         M   O         "O  RP         TP  Q         Q  R         R  S         S  U         U  LV         NV  ~W         W  X         X  Y         Y  [         [  F\         H\  x]         z]  ^         ^  _         _  a         a  @b         Bb  rc         tc  d         d  e         e  g         
g  :h         <h  li         ni  j         j  k         k  m         m  5n         6n  go         ho  p         p  q         q  r         r  /t         0t  au         bu  v         v  w         w  x         x  )z         *z  [{         \{  |         |  }         }  ~         ~  #         $  U         V                                              O         P                                              I         J  {                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               #!         *!  1         1  }=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   R      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-G;-x    G;-I-|    I-/}-\    /}-o-    o--h    -b-    b-x-	    x--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               B!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                               	  +	         ,	  N
         O
  q         q                                                           "          S  U                                                          F  ;          m  b              !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            )         *  S         T  }         ~                                   %         &  O	         P	  y
         z
                                   !         "  K         L  u         v                                              G         H  q         r                                              C         D  m         n               !         !  "          "  #         $  <%         >%  f&         h&  '         '  (         (  )         )  +         +  8,         :,  b-         d-  .         .  /         /  0         0  
2         2  43         63  ^4         `4  5         5  6         6  7         7  9         9  0:         2:  Z;         \;  <         <  =         =  >         >  @         @  ,A         .A  VB         XB  C         C  D         D  E         E  F          G  (H         :H  bI         dI  J         J  K         K  L         L  
N         N  4O         6O  ^P         `P  Q         Q  R         R  S         S  U         U  0V         2V  ZW         \W  X         X  Y         Y  Z         Z  \         \  ,]         .]  V^         X^  _         _  `         `  a         a  b          c  (d         *d  Re         Te  |f         ~f  g         g  h          Bm  8n          jn  `o          o  p          p  q         ʎ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ;!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-;-x    ;-;J-|    ;J-}-\    }--    --h    -Kb-    Kb---	    ---h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ?!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             "         #  E         F  h         i                       r  }                                    "          N  Y            {                                              !           [  }         ~                          !         !  	#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            1         2  c         d                                   +         ,  ]         ^  	         	  
         
             %         &  W         X                                               Q         R                                              K         L  }         ~                                   E          F   w!         x!  "         "  #          O6  ]7          7  8          8  9          9  ;          7;  E<         F  G         G  'I         (I  YJ         ZJ  K         K  L         L  M         M  !O         "O  SP         TP  Q         Q  R         R  S         S  U         U  MV         NV  W         W  X         X  Y         Y  [         [  G\         H\  y]         z]  ^         ^  _         _  a         a  Ab         Bb  sc         tc  d         d  e         e  	g         
g  ;h         <h  mi         ni  j         j  k         k  m         m  4n         6n  fo         ho  p         p  q         q  r         r  .t         0t  `u         bu  v         v  w         w  x         x  (z         *z  Z{         \{  |         |  }         }  ~         ~  "         $  T         V                                              N         P                                              H         J  z                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               %!         .!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   Q      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-G;-x    G;-I-|    I-/}-\    /}-o-    o--h    -b-    b-x-	    x--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               @!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                               	  +	         ,	  N
         O
  q         q                                                           !          S  T                                                          F  <          m  c              !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            (         *  R         T  |         ~                                   $         &  N	         P	  x
         z
                                             "  J         L  t         v                                              F         H  p         r                                              B         D  l         n               !         !  "         $  =%         >%  g&         h&  '         '  (         (  )         )  +         +  9,         :,  c-         d-  .         .  /         /  0         0  2         2  53         63  _4         `4  5         5  6         6  7         7  9         9  1:         2:  [;         \;  <         <  =         =  >         >  @         @  -A         .A  WB         XB  C         C  D         D  E         E  F          G  )H         :H  cI         dI  J         J  K         K  L         L  N         N  5O         6O  _P         `P  Q         Q  R         R  S         S  U         U  1V         2V  [W         \W  X         X  Y         Y  Z         Z  \         \  -]         .]  W^         X^  _         _  `         `  a         a  b          c  )d         *d  Se         Te  }f         ~f  g         g  h          q  r          
s   t          2t  (u          Zu  Pv          v  xw         ʎ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ?!         R!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     --    -:-x    :-{I-|    {I-|-\    |-
-    
--P    -ib-    ib-c-	    c-m-D
    m-3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               :!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P           5         6  X         Y  {	         |	  
         
           X  z         {                                                                     -            @  $          S  7          f  r                                     !          B!  N"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            .         0  ^         `                                               N         P  ~	         	  
         
                        >         @  n         p                                    .         0  ^         `                                               N         P  ~                                              >!         @!  n"         %  &         &  ((         *(  X)         Z)  *         *  +         +  ,         ,  .         .  H/         J/  x0         z0  1         1  2         2  4         
4  85         :5  h6         j6  7         7  8         8  9         9  (;         *;  X<         Z<  =         =  >         >  ?         ?  A         A  HB         JB  xC         zC  D         D  E         E  G         
G  8H          X^  V_          _  `          `  a          a  b           c  c         h   j         j  0k         2k  `l         bl  m         m  n         n  o         o   q         "q  Pr         Rr  s         s  t         t  u         u  w         w  @x         Bx  py         ry  z         z  {         {   }         }  0~         2~  `         b                                              L         N  {         }             ۈ         ݈             ;                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               2!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 }
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-4-x    4-C-|    C-v-\    v--    --h    -ob-    ob-x-	    x--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               B!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        7  =          n  t                                              &            9  	          L	  *
          _
  =         r                                                                                            B         C  e                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                 (  !          V  O            }         &  '         '  (         (  
*         *  6+         7+  b,         c,  -         -  .         .  /         /  1         1  >2         ?2  j3         k3  4         4  5         5  6         6  8         8  F9         G9  r:         s:  ;         ;  <         <  =         =  !?         #?  M@         O@  yA         {A  B         B  C         C  D         D  )F         +F  UG         WG  H         H  I         I  J         J  L         L  1M         3M  ]N         _N  O         O  P         P  Q         Q  S         S  9T         ;T  eU         gU  V         V  W         W  X         X  Z         Z  A[         C[  m\         o\  ]         ]  ^         ^  _         _  a         a  Ib         Kb  uc         wc  d         d  e         e  f         f  %h         'h  Qi         Si  }j         j  k         k  l          n  *o         ,o  Vp         Xp  q         q  r         r  s         s  u         u  2v         4v  ^w         `w  x         x  y         y  z         z  |         |  :}         <}  f~         h~                                              B         D  n         p             Ƈ         ȇ                         J         L  v         x             Ύ         Ў             &                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               6!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-y-     y--    -:-x    :-uI-|    uI-|-\    |-?-    ?--P    -ib-    ib-c-	    c-m-D
    m-3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               7!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P           5         6  X         Y  {	         |	  
         
           S  u         v                                                                     (            ;             N  3          a  l                                     !          =!  H"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            /         0  _         `                                               O         P  	         	  
         
                        ?         @  o         p                                    /         0  _         `                                               O         P                                                ?!         @!  o"         %  &         &  )(         *(  Y)         Z)  *         *  +         +  ,         ,  .         .  I/         J/  y0         z0  1         1  2         2  	4         
4  95         :5  i6         j6  7         7  8         8  9         9  );         *;  Y<         Z<  =         =  >         >  ?         ?  A         A  IB         JB  yC         zC  D         D  E         E  	G         
G  9H          *d  (e          Te  Rf          ~f  |g          g  h         h  j         j  1k         2k  al         bl  m         m  n         n  o         o  !q         "q  Qr         Rr  s         s  t         t  u         u  w         w  Ax         Bx  qy         ry  z         z  {         {  }         }  1~         2~  a         b                                              M         N  |         }             ܈         ݈             <          ʎ  ȏ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               2!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 }
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-4-x    4-C-|    C-v-\    v--    --h    -ob-    ob-x-	    x--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               @!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        7  <          n  s                                              &            9  	          L	  +
          _
  >         r                                                                                           =         >  `                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                             <  5          j  c         &  '         '  (         (  	*         *  5+         7+  a,         c,  -         -  .         .  /         /  1         1  =2         ?2  i3         k3  4         4  5         5  6         6  8         8  E9         G9  q:         s:  ;         ;  <         <  =         =  "?         #?  N@         O@  zA         {A  B         B  C         C  D         D  *F         +F  VG         WG  H         H  I         I  J         J  L         L  2M         3M  ^N         _N  O         O  P         P  Q         Q  S         S  :T         ;T  fU         gU  V         V  W         W  X         X  Z         Z  B[         C[  n\         o\  ]         ]  ^         ^  _         _  a         a  Jb         Kb  vc         wc  d         d  e         e  f         f  &h         'h  Ri         Si  ~j         j  k         k  l          n  +o         ,o  Wp         Xp  q         q  r         r  s         s  u         u  3v         4v  _w         `w  x         x  y         y  z         z  |         |  ;}         <}  g~         h~                                              C         D  o         p             Ǉ         ȇ                         K         L  w         x             ώ         Ў             '                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               8!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,-m-     m--    -;-    ;-5J-    5J-}-l    }-!-    !--P    --    -b-h	    b--p
    -#-
    #--    -3׳-L                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               5!         >!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             "         #  E         F  h         i                       r  ~                                    #          N  Z            z                                              !           [  }         ~                          !         !  	#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        %  &         &  '         '  )         
)  6*         8*  d+         f+  ,         ,  -         -  .         .  0         0  J1         L1  x2         z2  3         3  4         4  6         6  07         27  ^8         `8  9         9  :         :  ;         ;  =         =  D>         F>  r?         t?  @         @  A         A  B         B  *D         ,D  XE         ZE  F         F  G          G  H          H  I          J  K          @K  ?L         nL  M         M  N         N  O         O  %Q         &Q  SR         TR  S         S  T         T  U         U  W         W  9X         :X  gY         hY  Z         Z  [         [  \         \  ^          ^  M_         N_  {`         |`  a         a  b         b  d         d  3e         4e  af         bf  g         g  h         h  i         i  k         k  Gl         Hl  um         vm  n         n  o         o  p           q  q         .r  Zs         \s  t         t  u         u  v         v  x         x  @y         By  nz         pz  {         {  |         |  }         }  &         (  T         V                        ރ                      :         <  h         j             ĉ         Ɖ                       "  N         P  |         ~                                b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ?!         V!  1         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    j
         k
             @                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     P
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-m-     m--    -A;-x    A;-I-|    I-)}-\    )}-,-    ,--h    -ub-    ub--	    --h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               0!         @!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                               	  +	         ,	  N
         O
  q         q                                                           "          S  U                                                          F  ;          m  b              !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            -         .  [         \                                              A         B  o	         p	  
         
                        '         (  U         V                                              ;         <  i         j                                   !         "  O         P  }         ~                        !         !  5"         6"  c#         %  &         &  '         '  %)         &)  S*         T*  +         +  ,         ,  -         -  /         /  90         :0  g1         h1  2         2  3         3  4         4  6          6  M7         N7  {8         |8  9         9  :         :  <         <  3=         4=  a>         b>  ?         ?  @         @  A         A  C         C  GD         HD  uE         vE  F         F  G         G  H         TI  J         J  K          d  e          4e  0f          bf  ^g          g  h          h  i           ?         A  m         o             ɐ         ː             %         '  S         U                        ݗ         ߗ             9         ;  g         i             Ý         ŝ                      !  M         O  {         }             פ         ٤             3         5  a         c                                                      >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               9!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,-m-     m--    -;-    ;-5J-    5J-}-l    }-!-    !--P    --    -b-h	    b--p
    -#-
    #--    -3׳-L                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               3!         >!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             "         #  E         F  h         i                       r  }                                    "          N  Y            {                                              !           [  }         ~                          !         !  	#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        %  &         &  '         '  	)         
)  7*         8*  e+         f+  ,         ,  -         -  .         .  0         0  K1         L1  y2         z2  3         3  4         4  6         6  17         27  _8         `8  9         9  :         :  ;         ;  =         =  E>         F>  s?         t?  @         @  A         A  B         B  +D         ,D  YE         ZE  F         F  G         G  H         H  J         J  ?K         @K  mL         nL  M         M  N         N  O         O  $Q         &Q  RR         TR  S         S  T         T  U         U  
W         W  8X         :X  fY         hY  Z         Z  [         [  \         \  ^          ^  L_         N_  z`         |`  a         a  b         b  d         d  2e         4e  `f         bf  g         g  h         h  i         i  k         k  Fl         Hl  tm         vm  n         n  o         o  p          q  -r          .r  -s          \s  [t          t  u          u  v          v  w         x  Ay         By  oz         pz  {         {  |         |  }         }  '         (  U         V                        ߃                      ;         <  i         j             ŉ         Ɖ             !         "  O         P  }         ~                                b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ?!         V!  1         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    j
         k
             @                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     Q
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-s-     s--    -A;-x    A;-I-|    I-)}-\    )}-,-    ,--h    -ib-    ib--	    --h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               4!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                               	  +	         ,	  N
         O
  q         q                                                           !          S  T                                                          F  <          m  c              !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ,         .  Z         \                                              @         B  n	         p	  
         
                        &         (  T         V                                              :         <  h         j                                             "  N         P  |         ~                        !         !  4"         6"  b#         %  &         &  '         '  $)         &)  R*         T*  +         +  ,         ,  -         -  
/         /  80         :0  f1         h1  2         2  3         3  4         4  6          6  L7         N7  z8         |8  9         9  :         :  <         <  2=         4=  `>         b>  ?         ?  @         @  A         A  C         C  FD         HD  tE         vE  F         F  G         G  H         TI  J         J  K          i  j          k  l          Hl  Dm          n  o          o  p           @         A  n         o             ʐ         ː             &         '  T         U                        ޗ         ߗ             :         ;  h         i             ĝ         ŝ                       !  N         O  |         }             ؤ         ٤             4         5  b         c                                                      >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               5!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,-m-     m--    -5-    5-GD-    GD-w-l    w-4-    4--x    -cb-	    cb--
    --
    -m-|    m-ճ-    ճ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               3!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         '            N  C          u  j                                              )  %	          \	  X
          
                                   *         +  M         N  p         p                                                                            b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         l  {                                  &  5          d  s         D%  n&         JL  uM         vM  N         N  O         O  P         P  %R         &R  QS         RS  }T         ~T  U         U  V         V  X         X  -Y         .Y  YZ         ZZ  [         [  \         \  ]         ]  	_         
_  5`         6`  aa         ba  b         b  c         c  d         d  f         f  =g         >g  ih         jh  i         i  j         j  k         k  m         m  En         Fn  qo         ro  p         p  q         q  r         r  !t         "t  Mu         Nu  yv         zv  w         w  x         x  y         y  ){         *{  U|         V|  }         }  ~         ~                        1         2  ]         ^                                              9         :  e         f                                              A         B  m         n             ē         Ɠ                        H         J  t         v             ̚         Κ             $         &  P         R  |         ~             ԡ         ֡              ,         .  X         Z                        ܨ         ި           
  4         6  `         b                                              <         >  h         j                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               4!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    j
         k
             @      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    j
         k
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            0     -,-X     ,--     -/-    /-;-x    ;-/J-|    /J-}}-\    }}-u-    u--h    -Qb-    Qb-낳-	    낳--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               F!         V!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             "         #  E         F  h         i                                                           7  ,          ^  S                                              *  0          a  g         [  }         ~                          !         !  	#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            .         0  ^         `                                               N         P  ~	         	  
         
                        >         @  n         p                                    .         0  ^         `                                               N         P  ~                                              >!         @!  n"         p"  #         D%  s&         t&  '         '  (         (  *         *  3+         4+  c,         d,  -         -  .         .  /         /  #1         $1  S2         T2  3         3  4         4  5         5  7         7  C8         D8  s9         t9  :         :  ;         ;  =         =  3>         4>  c?         d?  @         @  A         A  B         B  #D         $D  SE         TE  F         F  G         G  H          JL  [M          M  N          N  O          P  $Q          VQ  gR         ir  s         s  t         t  u         u  'w         )w  Wx         Yx  y         y  z         z  {         {  }         }  G~         I~  w         y             ׁ         ف           	  7         9  g         i             Ǉ         ɇ             '         )  W         Y                                              G         I  w         y             ה         ٔ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               0!         :!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,-g-     g--    -5-    5-GD-    GD-w-l    w-4-    4--x    -]b-	    ]b--
    --
    -m-|    m-ճ-    ճ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               1!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         '            N  D          u  k                                              )  $	          \	  W
          
                                   *         +  M         N  p         p                                                                            b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                   -          \  k                     D%  o&         JL  tM         vM  N         N  O         O  P         P  $R         &R  PS         RS  |T         ~T  U         U  V         V   X         X  ,Y         .Y  XZ         ZZ  [         [  \         \  ]         ]  _         
_  4`         6`  `a         ba  b         b  c         c  d         d  f         f  <g         >g  hh         jh  i         i  j         j  k         k  m         m  Dn         Fn  po         ro  p         p  q         q  r         r   t         "t  Lu         Nu  xv         zv  w         w  x         x  y         y  ({         *{  T|         V|  }         }  ~         ~                        0         2  \         ^                                              8         :  d         f                                              @         B  l         n             œ         Ɠ                        I         J  u         v             ͚         Κ             %         &  Q         R  }         ~             ա         ֡             -         .  Y         Z                        ݨ         ި  	         
  5         6  a         b                                              =         >  i         j                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               2!         @!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    j
         k
             @      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    j
         k
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            0     -,-X     ,--     -/-    /-;-x    ;-/J-|    /J-}}-\    }}-{-    {--h    -Qb-    Qb-낳-	    낳--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               F!         V!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             "         #  E         F  h         i                                                           7  -          ^  T                                              *  /          a  f         [  }         ~                          !         !  	#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            /         0  _         `                                               O         P  	         	  
         
                        ?         @  o         p                                    /         0  _         `                                               O         P                                                ?!         @!  o"         p"  #         D%  r&         t&  '         '  (         (  *         *  2+         4+  b,         d,  -         -  .         .  /         /  "1         $1  R2         T2  3         3  4         4  5         5  7         7  B8         D8  r9         t9  :         :  ;         ;  =         =  2>         4>  b?         d?  @         @  A         A  B         B  "D         $D  RE         TE  F         F  G         G  H          R  S          S  T          U  0V          bV  sW          W  X         ir  s         s  t         t  u         u  (w         )w  Xx         Yx  y         y  z         z  {         {  }         }  H~         I~  x         y             ؁         ف           	  8         9  h         i             ȇ         ɇ             (         )  X         Y                                              H         I  x         y             ؔ         ٔ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               .!         :!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-m-     m--    -;-x    ;-)J-|    )J-w}-\    w}-96-    96--h    -b-    b-m-	    m--h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               4!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             "         #  E         F  h         i                       r  ~                                    #          N  Z            z                                              !           [  }         ~                          !         !  	#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            (         *  R         T  |         ~                                   $         &  N	         P	  x
         z
                                             "  J         L  t         v                                              F         H  p         r                                              B         D  l         n               !         !  "          ?  '@          Y@  jA          A  B          B  C          "D  3E         K  L         L  N         N  1O         2O  [P         \P  Q         Q  R         R  S         S  U         U  -V         .V  WW         XW  X         X  Y         Y  Z         Z  [          \  )]         *]  S^         T^  }_         ~_  `         `  a         a  b         b  %d         &d  Oe         Pe  yf         zf  g         g  h         h  i         i  !k         "k  Kl         Ll  um         vm  n         n  o           6         8  `         b                        ޚ                    
  2         4  \         ^                        ڡ         ܡ             .         0  X         Z                        ֨         ب              *         ,  T         V  ~                      ү         ԯ             &         (  P         R  z         |             ζ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               M!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     0
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -A-    A-;;-    ;;-I-    I-#}-d    #}--    -U-0    U--    -b-H	    b-c-P
    c-C-
    C-'׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               W!         X!  2         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                               	  +	         ,	  N
         O
  q         q                                                           "          S  U                                                          F  ;          m  b              !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q   w  operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       \            1         2  c         d                                   +         ,  ]         ^  	         	  
         
             %         &  W         X                                               Q         R                                              K         L  }         ~                                   E          F   w!         x!  "         "  #         #  %         %  ?&         @&  q'         r'  (         (  )         )  +         +  9,         :,  k-         l-  .         .  /         /  1         1  32         42  e3         f3  4         4  5         5  6         6  -8         .8  _9         `9  :         :  ;         ;  <         <  '>         (>  Y?         Z?  @         @  A         A  B         B  !D         "D  SE         TE  F         F  G         G  H         H  J         ir  s         s  t         t  u         u  /w         1w  ax         cx  y          y  z          z  {          {  |          +}  .~          ]~  `                                 #         %  U         W                                              O         Q                                              I                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               W!         X!  2         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 q
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-s-     s--    -;-x    ;-)J-|    )J-w}-\    w}-96-    96--h    -b-    b-m-	    m--h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               6!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             "         #  E         F  h         i                       r  }                                    "          N  Y            {                                              !           [  }         ~                          !         !  	#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            )         *  S         T  }         ~                                   %         &  O	         P	  y
         z
                                   !         "  K         L  u         v                                              G         H  q         r                                              C         D  m         n               !         !  "          eE  vF          F  G          G  H          .I  ?J          qJ  K         K  L         L  N         N  0O         2O  ZP         \P  Q         Q  R         R  S         S  U         U  ,V         .V  VW         XW  X         X  Y         Y  Z         Z  [          \  (]         *]  R^         T^  |_         ~_  `         `  a         a  b         b  $d         &d  Ne         Pe  xf         zf  g         g  h         h  i         i   k         "k  Jl         Ll  tm         vm  n         n  o           7         8  a         b                        ߚ           	         
  3         4  ]         ^                        ۡ         ܡ             /         0  Y         Z                        ר         ب             +         ,  U         V                        ӯ         ԯ             '         (  Q         R  {         |             ϶                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               M!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     /
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -;-    ;-;;-    ;;-I-    I-#}-d    #}--    -U-0    U--    -b-H	    b-c-P
    c-C-
    C-'׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               T!         V!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                               	  +	         ,	  N
         O
  q         q                                                           !          S  T                                                          F  <          m  c              !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q   w  operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       \            0         2  b         d                                   *         ,  \         ^  	         	  
         
             $         &  V         X                                               P         R                                              J         L  |         ~                                   D          F   v!         x!  "         "  #         #  %         %  >&         @&  p'         r'  (         (  )         )  +         +  8,         :,  j-         l-  .         .  /         /   1         1  22         42  d3         f3  4         4  5         5  6         6  ,8         .8  ^9         `9  :         :  ;         ;  <         <  &>         (>  X?         Z?  @         @  A         A  B         B   D         "D  RE         TE  F         F  G         G  H         H  J         ir  s         s  t         t  u         u  0w         1w  bx         cx  y         y  z         z  {         {  *}         +}  \~         ]~                          ā                      %  (          W  Z                                            P         Q                                              J                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               S!         T!  1         1  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 q
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-U-     U--    -5-x    5-YD-|    YD-w-\    w-+-    +--h    -b-    b-m-	    m--h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     f            !          X!  1          2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         %            J  ;          o  `                                              #  '	          X	  \
          
                                   0         1  U         V  z                                                       <                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            4         6  j         l                                   B         D  x         z  	         	  
         
             P         R                                   (         *  ^         `                                    6         8  l         n                                   D         F  z         |               !         !  #         #  R$          $  %          %  &           '  (          8(  <)          p)  t*          I  UJ         VJ  K         K  L         L  M         M  -O         .O  cP         dP  Q         Q  R         R  T         T  ;U         <U  qV         rV  W         W  X         X  Z         Z  I[         J[  \         \  ]         ]  ^         ^  !`         "`  Wa         Xa  b         b  c         c  d         d  /f         0f  eg         fg  h         h  i         i  k         k  =l         >l  sm         tm  n         n  o         o  q         q  Kr         Lr  s         s  t         t  u         u  #w         $w  Yx         Zx  y         y  z         z  {         {  1}         2}  g~         h~             Ӏ         Ԁ  	         
  ?         @  u         v                                   M         N                                   %         &  [         \             Ǒ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               W!         X!  2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     +
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            4     -,-`     ,-=-     =--    -w;-    w;-J-    J-_}-d    _}-a-    a--p    -ob-    ob-	-
    	--
    --t    --                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     g            
!          F!  1          1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             $         %  I         J  n         o                                                           1  "          V  G          {                                      $          O  Y         =  a         b                          !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       8  ?          p  w                                 $  %         %  &         &  (         (  ')         ()  M*         N*  s+         t+  ,         ,  -         -  .         .  0         0  11         21  W2         X2  }3         ~3  4         4  5         5  6         6  8         8  ;9         <9  a:         b:  ;         ;  <         <  =         =  >         >  @          @  EA         FA  kB         lB  C         C  D         D  E         E  G         G  (H         *H  NI         PI  tJ         vJ  K         K  L         L  M         M  O         O  2P         4P  XQ         ZQ  ~R         R  S         S  T         T  U         U  W         W  <X         >X  bY         dY  Z         Z  [         [  \         \  ]         ]   _         "_  F`         H`  la         na  b         b  c         c  d         d  f         
f  0g         2g  Xh         Zh  i         i  j         j  k         k  l         l   n         "n  Ho         Jo  np         pp  q         q  r         r  s         s  u         u  ,v         .v  Rw         Tw  xx         zx  y         y  z         z  {         {  }         }  6~         8~  \         ^                        ΂         Ђ                        @         B  f         h                        ؊         ڊ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               I!         J!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    l
         m
             F      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     '
                   
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,-O-     O--    -5-x    5-YD-|    YD-w-\    w-+-    +--h    -b-    b-m-	    m--h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     e            !          T!  1          1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         %            J  <          o  a                                              #  &	          X	  [
          
                                   0         1  U         V  z                                                       <                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            5         6  k         l                                   C         D  y         z  	         	  
         
             Q         R                                   )         *  _         `                                   7         8  m         n                                   E         F  {         |               !         !  #         #  S$          *  +          +  ,          -  .          P.  T/          /  0          I  TJ         VJ  K         K  L         L  M         M  ,O         .O  bP         dP  Q         Q  R         R  T         T  :U         <U  pV         rV  W         W  X         X  Z         Z  H[         J[  ~\         \  ]         ]  ^         ^   `         "`  Va         Xa  b         b  c         c  d         d  .f         0f  dg         fg  h         h  i         i  k         k  <l         >l  rm         tm  n         n  o         o  q         q  Jr         Lr  s         s  t         t  u         u  "w         $w  Xx         Zx  y         y  z         z  {         {  0}         2}  f~         h~             Ҁ         Ԁ           
  >         @  t         v                                   L         N                                   $         &  Z         \             Ƒ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               S!         T!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     *
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            4     -,-`     ,-C-     C--    -w;-    w;-J-    J-_}-d    _}-g-    g--p    -ob-    ob-	-
    	--
    --t    --                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     h            !          J!  1          1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             $         %  I         J  n         o                                                           1  #          V  H          {                                      #          O  X         =  a         b                          !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                     P  W            	          	  
          
           $  %         %  &         &   (         (  &)         ()  L*         N*  r+         t+  ,         ,  -         -  .         .  
0         0  01         21  V2         X2  |3         ~3  4         4  5         5  6         6  8         8  :9         <9  `:         b:  ;         ;  <         <  =         =  >         >  @          @  DA         FA  jB         lB  C         C  D         D  E         E  G         G  )H         *H  OI         PI  uJ         vJ  K         K  L         L  M         M  O         O  3P         4P  YQ         ZQ  R         R  S         S  T         T  U         U  W         W  =X         >X  cY         dY  Z         Z  [         [  \         \  ]         ]  !_         "_  G`         H`  ma         na  b         b  c         c  d         d  	f         
f  1g         2g  Yh         Zh  i         i  j         j  k         k  l         l  !n         "n  Io         Jo  op         pp  q         q  r         r  s         s  u         u  -v         .v  Sw         Tw  yx         zx  y         y  z         z  {         {  }         }  7~         8~  ]         ^                        ς         Ђ                        A         B  g         h                        ي         ڊ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               I!         J!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    l
         m
             F      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     (
                   
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,-[-     [--    -:-x    :-cI-|    cI-|-\    |--    --h    -b-    b-kx-	    kx--h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     f            !          X!  1          2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         	  -         .  R         S  w	         x	  
         
           N  r         s                                                                     1            F  "          [  7          p  p                                     !          D!  D"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            2         4  f         h                                   6         8  j         l  	         	  
         
             :         <  n         p                        
           >         @  r         t                                   B         D  v         x                                   F         H  z          |   !         !  "         "  $         $  J%         L%  ~&         &  '         '  (         (  *         *  N+         P+  ,         ,  -         -  .         .  0          0  R1         T1  2         2  3         3  4         4  "6         $6  V7         X7  8         8  9         9  :         :  &<         (<  Z=         \=  >         >  ?         ?  @         @  *B         ,B  ^C         `C  D         D  E         E  F         F  .H         ^H  I         I  J         J  K         K  -M         .M  aN         bN  O         O  P         P  Q         Q  1S         2S  eT         fT  U         U  V         V  X         X  5Y         6Y  iZ         jZ  [         [  \         \  ^         ^  9_         :_  m`         n`  a         a  b         b  	d         
d  =e         >e  qf         rf  g         g  h         h  j         j  Ak         Bk  ul          z  {          :{  %|          ^|  I}          }  m~          ~                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               Z!         ^!  	2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 k
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            <     -,-p     ,-s-     s--    -4-    4-C-    C-v-t    v---     --C5-X    C5--    -9b-p	    9b--x
    --
    -m-    m-ճ-\    ճ--<                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               9!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P              
          5  ?          j  t                                  	                        3  	          H	  .
          ]
  C         r                                              *                                  %         &  J         K  o                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        V"  {#         }#  $         H  I         I  J         J  K         K  M         M  EN         GN  mO         oO  P         P  Q         Q  R         R  T         T  5U         7U  ]V         _V  W         W  X         X  Y         Y  Z         Z  %\         '\  M]         O]  u^         w^  _         _  `         `  a         a  c         c  =d         ?d  ee         ge  f         f  g         g  h         h  j         j  -k         /k  Ul         Wl  }m         m  n         n  o         o  p         p  r         r  Es         Gs  mt         ot  u         u  v         v  w         w  y         y  5z         7z  ]{         _{  |         |  }         }  ~         ~             %         '  M         O  u         w             Ņ         ǅ                        =         ?  e         g                        ݍ         ߍ             ,         .  S         U  {         }             ˔         ͔                                   E  <          m  d                                              3         5  [         ]                        ӣ         գ             #         %  K         M  s         u             ê         Ū                        ;                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     g            !          L!  1          1  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    l
         m
             F                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    n
         o
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            0     -,-X     ,-U-     U--    -:-x    :-]I-|    ]I-|-\    |--    --h    -b-    b-kx-	    kx--h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     e            !          T!  1          1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         	  -         .  R         S  w	         x	  
         
           I  m         n                                                                     ,  	          A            V  3          k  j                                  
   	!          ?!  >"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            3         4  g         h                                   7         8  k         l  	         	  
         
             ;         <  o         p                                   ?         @  s         t                                   C         D  w         x                                   G         H  {          |   !         !  "         "  $         $  K%         L%  &         &  '         '  (         (  *         *  O+         P+  ,         ,  -         -  .         .  0          0  S1         T1  2         2  3         3  4         4  #6         $6  W7         X7  8         8  9         9  :         :  '<         (<  [=         \=  >         >  ?         ?  @         @  +B         ,B  _C         `C  D         D  E         E  F         F  /H         ^H  I         I  J         J  K         K  ,M         .M  `N         bN  O         O  P         P  Q         Q  0S         2S  dT         fT  U         U  V         V   X         X  4Y         6Y  hZ         jZ  [         [  \         \  ^         ^  8_         :_  l`         n`  a         a  b         b  d         
d  <e         >e  pf         rf  g         g  h         h  j         j  @k         Bk  tl                        ف                      6  !          Z  E                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               X!         ^!  2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 k
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            <     -,-p     ,-y-     y--    -4-    4-C-    C-v-t    v---     --C5-X    C5--    -?b-p	    ?b--x
    --
    -s-    s-ճ-\    ճ--<                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ;!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P              	          5  >          j  s                                  	                        3  	          H	  /
          ]
  D         r                                              *                                           !  E         F  j                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        V"  |#         }#  $         H  I         I  J         J  K         K  M         M  FN         GN  nO         oO  P         P  Q         Q  R         R  T         T  6U         7U  ^V         _V  W         W  X         X  Y         Y  Z         Z  &\         '\  N]         O]  v^         w^  _         _  `         `  a         a  c         c  >d         ?d  fe         ge  f         f  g         g  h         h  j         j  .k         /k  Vl         Wl  ~m         m  n         n  o         o  p         p  r         r  Fs         Gs  nt         ot  u         u  v         v  w         w  y         y  6z         7z  ^{         _{  |         |  }         }  ~         ~             &         '  N         O  v         w             ƅ         ǅ                        >         ?  f         g                        ލ         ߍ             -         .  T         U  |         }             ̔         ͔                        D         E  l         m                                    ܝ                      5  ,          ]  T            |           ԣ         գ             $         %  L         M  t         u             Ī         Ū                        <                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     h            !          L!  1          1  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    l
         m
             F                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    o
         p
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            0     -,-X     ,--     -/-    /-5-x    5-_D-|    _D-w-\    w--    --h    -Kb-    Kb-kx-	    kx--h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               C!         X!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         %            J  ;          o  `                                              #  '	          X	  \
          
                                   0         1  U         V  z                                                       <                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |           l                                                          $  %         %  '         '  8(         :(  j)         l)  *         *  +         +   -         -  2.         4.  d/         f/  0         0  1         1  2         2  ,4         .4  ^5         `5  6         6  7         7  8         8  &:         (:  X;         Z;  <         <  =         =  >         >   @         "@  RA         TA  B         B  C         C  D         D  F         F  LG         NG  ~H         H  I         I  K         K  BL         DL  tM         vM  N         N  O         O  
Q         Q  <R         >R  nS         pS  T         T  U         U  W         W  7X         8X  iY         jY  Z         Z  [         [  \          ]  1^         2^  c_         d_  `         `  a         a  b         b  +d         ,d  ]e         ^e  f         f  g         g  h         h  %j         &j  Wk         Xk  l         l  m         m  n         n  p          p  Qq         Rq  r         r  s         s  t         t  v         v  Kw         Lw  }x         ~x  y         y  z         z  |         |  E}         F}  w~         x~             ۀ         ܀             ?         @  q         r             Ն         ֆ             9         :  k         l             ό         Ќ             3         d                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     f            !          T!  1          1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 k
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-y-     y--    -};-x    };-J-|    J-e}-\    e}-.-    .--h    -Kb-    Kb-n-	    n--h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               :!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             $         %  I         J  n         o                                                           1  "          V  G          {                                      $          O  Y         =  a         b                          !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            (         *  R         T  |         ~                                   $         &  N	         P	  x
         z
                                             "  J         L  t         v                                              F         H  p         r                                              B         D  l         n               !          ~8  p9          9  :          :  ;          ;  <          
=  =         @G  iH         jH  I         I  J         J  K         K  "M         $M  LN         NN  vO         xO  P         P  Q         Q  R         R  T          T  HU         JU  rV         tV  W         W  X         X  Y         Y  [         [  D\         F\  n]         p]  ^         ^  _         _  `         `  b         b  @c         Bc  jd         ld  e         e  f         f  g         g  i         i  <j         >j  fk         hk  l         l  m           ͏         Ώ             !         "  K         L  u         v             ɖ         ʖ                        G         H  q         r             ŝ         Ɲ                        C         D  m         n                      ¤                        ?         @  i         j                                              ;                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     g            !          R!  1          1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     O
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-5-x    5-_D-|    _D-w-\    w--    --h    -Eb-    Eb-kx-	    kx--h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               A!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         %            J  <          o  a                                              #  &	          X	  [
          
                                   0         1  U         V  z                                                       <                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                      @  ,          `  L            l              !         $  %         %  '         '  9(         :(  k)         l)  *         *  +         +  -         -  3.         4.  e/         f/  0         0  1         1  2         2  -4         .4  _5         `5  6         6  7         7  8         8  ':         (:  Y;         Z;  <         <  =         =  >         >  !@         "@  SA         TA  B         B  C         C  D         D  F         F  MG         NG  H         H  I         I  K         K  CL         DL  uM         vM  N         N  O         O  Q         Q  =R         >R  oS         pS  T         T  U         U  W         W  6X         8X  hY         jY  Z         Z  [         [  \          ]  0^         2^  b_         d_  `         `  a         a  b         b  *d         ,d  \e         ^e  f         f  g         g  h         h  $j         &j  Vk         Xk  l         l  m         m  n         n  p          p  Pq         Rq  r         r  s         s  t         t  v         v  Jw         Lw  |x         ~x  y         y  z         z  |         |  D}         F}  v~         x~             ڀ         ܀             >         @  p         r             Ԇ         ֆ             8         :  j         l             Ό         Ќ              2         d                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     e            !          P!  1          1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 k
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-y-     y--    -};-x    };-J-|    J-e}-\    e}-.-    .--h    -Qb-    Qb-n-	    n--h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               :!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             $         %  I         J  n         o                                                           1  #          V  H          {                                      #          O  X         =  a         b                          !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            )         *  S         T  }         ~                                   %         &  O	         P	  y
         z
                                   !         "  K         L  u         v                                              G         H  q         r                                              C         D  m         n               !          ->  ?          P?  B@          s@  eA          A  B          B  C         @G  hH         jH  I         I  J         J  K         K  #M         $M  MN         NN  wO         xO  P         P  Q         Q  R         R  T          T  IU         JU  sV         tV  W         W  X         X  Y         Y  [         [  E\         F\  o]         p]  ^         ^  _         _  `         `  b         b  Ac         Bc  kd         ld  e         e  f         f  g         g  i         i  =j         >j  gk         hk  l         l  m           ̏         Ώ                       "  J         L  t         v             Ȗ         ʖ                        F         H  p         r             ĝ         Ɲ                        B         D  l         n                      ¤                        >         @  h         j                                              :                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     h            !          V!  1          2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     N
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-:-x    :-iI-|    iI-|-\    |-1-    1--h    -3b-    3b--	    --h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ;!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         	  -         .  R         S  w	         x	  
         
           N  r         s                                                                     1            F  "          [  7          p  p                                     !          D!  D"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            .         0  ^         `                                               N         P  ~	         	  
         
                        >         @  n         p                                    .         0  ^         `                                               N         P  ~                                              >!         @!  n"         p"  #         #  $         $  %          &  .'         0'  ^(         `(  )         )  *         *  +         +  -          -  N.         P.  ~/         /  0         0  1         1  3         3  >4         @4  n5         p5  6         6  7         7  8          9  .:         0:  ^;         `;  <         <  =         =  >         >  @          @  NA         PA  ~B         B  C         C  D         D  F         F  >G         @G  nH         pH  I         I  J          K  M          :M  @N          vN  |O          O  P          P  Q         d                      Ĕ             "         $  R         T                                              B         D  r         t             ҡ         ԡ             2         4  b         d             §         ħ             "         $  R         T                                              B         D  r                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     [                       H!  1          1  w=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   B      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-y-     y--    -4-x    4-C-|    C-v-\    v-	-    	--h    -ib-    ib-!n-	    !n--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               8!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P              
          5  ?          j  t                                  	                        3  	          H	  .
          ]
  C         r                                              *                                  %         &  J         K  o                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            *         ,  V         X                                              2         4  ^	         `	  
         
                                   :         <  f         h                                              B         D  n         p                                               J         L  v         x                          !         !  &#          d#  x$          $  %          %  &          $'  8(          d(  x)         HI  rJ         tJ  K         K  L         L  M         M  "O         $O  NP         PP  zQ         |Q  R         R  S         S  T          U  *V         ,V  VW         XW  X         X  Y         Y  Z         Z  \         \  2]         4]  ^^         `^  _         _  `         `  a         a  c         c  :d         <d  fe         he  f         f  g         g  h         h  j         j  Bk         Dk  nl         pl  m         m  n         n  o         o  q          q  Jr         Lr  vs         xs  t         t  u         u  v         v  &x         (x  Ry         Ty  ~z         z  {         {  |         |  ~         ~  .         0  Z         \                        ރ           
           6         8  b         d                                              >         @  j         l                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               8!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     O
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-:-x    :-cI-|    cI-|-\    |-1-    1--h    -3b-    3b--	    --h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               =!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         	  -         .  R         S  w	         x	  
         
           I  m         n                                                                     ,  	          A            V  3          k  j                                  
   	!          ?!  >"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            /         0  _         `                                               O         P  	         	  
         
                        ?         @  o         p                                    /         0  _         `                                               O         P                                                ?!         @!  o"         p"  #         #  $         $  %          &  /'         0'  _(         `(  )         )  *         *  +         +  -          -  O.         P.  /         /  0         0  1         1  3         3  ?4         @4  o5         p5  6         6  7         7  8          9  /:         0:  _;         `;  <         <  =         =  >         >  @          @  OA         PA  B         B  C         C  D         D  F         F  ?G         @G  oH         pH  I         I  J          *R  0S          fS  lT          T  U          U  V          W   X         d             Ô         Ĕ             #         $  S         T                                              C         D  s         t             ӡ         ԡ             3         4  c         d             ç         ħ             #         $  S         T                                              C         D  s                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     \             !          H!  1          1  x=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   A      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-y-     y--    -4-x    4-C-|    C-v-\    v-	-    	--h    -ib-    ib-!n-	    !n--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               8!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P              	          5  >          j  s                                  	                        3  	          H	  /
          ]
  D         r                                              *                                           !  E         F  j                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            +         ,  W         X                                              3         4  _	         `	  
         
                                   ;         <  g         h                                              C         D  o         p                                               K         L  w         x                          !         !  '#          )  *          *  +          $,  8-          d-  x.          .  /         HI  sJ         tJ  K         K  L         L  M         M  #O         $O  OP         PP  {Q         |Q  R         R  S         S  T          U  +V         ,V  WW         XW  X         X  Y         Y  Z         Z  \         \  3]         4]  _^         `^  _         _  `         `  a         a  c         c  ;d         <d  ge         he  f         f  g         g  h         h  j         j  Ck         Dk  ol         pl  m         m  n         n  o         o  q          q  Kr         Lr  ws         xs  t         t  u         u  v         v  'x         (x  Sy         Ty  z         z  {         {  |         |  ~         ~  /         0  [         \                        ߃                      7         8  c         d                                              ?         @  k         l                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               8!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     N
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-;-x    ;-#J-|    #J-q}-\    q}--    --h    -Kb-    Kb--	    --h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               C!         V!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             $         %  I         J  n         o                       r  r                                              F  F          {  l                                                          =  a         b                          !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ,         .  Z         \                                              @         B  n	         p	  
         
                        &         (  T         V                                              :         <  h         j                                             "  N         P  |         ~                        !         !  4"         6"  b#         d#  $         $  %         %  &         &  (         (  H)         J)  v*         x*  +         +  ,         ,   .         .  ./         0/  \0         ^0  1         1  2         2  3         3  5         5  B6         D6  p7         r7  8         8  9         9  :         :  (<         *<  V=         X=  >         >  ?         ?  @         @  B         B  <C         >C  jD         lD  E         E  F           U  U          ,V  "W          XW  NX          X  zY          Y  Z         l  m         m  o          o  Mp         Np  {q         |q  r         r  s         s  u         u  3v         4v  aw         bw  x         x  y         y  z         z  |         |  G}         H}  u~         v~             р         Ҁ              -         .  [         \                                              A         B  o         p             ˍ         ̍             '                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               '!         6!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   B      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-S;-x    S;-I-|    I-;}-\    ;}-m-    m--h    -ob-    ob-x-	    x--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               <!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 '	         (	  L
         M
  q         {                                              3          4  .          i  c                                              =  .          b  S            x              !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            -         .  [         \                                              A         B  o	         p	  
         
                        '         (  U         V                                              ;         <  i         j                                   !         "  O         P  }         ~                        !         !  5"         6"  c#         d#  $         $  %         %  &         &  (         (  H)         J)  v*         x*  +         +  ,         ,   .         .  ./         0/  \0         ^0  1         1  2         2  3         3  5         5  B6         D6  p7         r7  8         8  9         9  :         :  (<         *<  V=         X=  >         >  ?         ?  @         @  B         B  <C         >C  jD         lD  E         E  F         F  G         G  "I         $I  PJ         RJ  ~K         K  L         L  M         M  O         
O  6P         8P  dQ         fQ  R         R  S         S  T         T  V         V  JW         LW  xX         zX  Y         Y  Z         Z  \         \  0]         2]  ^^         `^  _         _  `         `  a         a  c         c  Dd         Fd  re         te  f         f  g         g  h          (x  y          Ty  Dz          z  p{          {  |          |  }         ʐ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               9!         J!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-;-x    ;-#J-|    #J-q}-\    q}--    --h    -Qb-    Qb--	    --h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               A!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             $         %  I         J  n         o                       r  q                                              F  E          {  m                                                         =  a         b                          !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            -         .  [         \                                              A         B  o	         p	  
         
                        '         (  U         V                                              ;         <  i         j                                   !         "  O         P  }         ~                        !         !  5"         6"  c#         d#  $         $  %         %  &         &  (         (  I)         J)  w*         x*  +         +  ,         ,  .         .  //         0/  ]0         ^0  1         1  2         2  3         3  5         5  C6         D6  q7         r7  8         8  9         9  :         :  )<         *<  W=         X=  >         >  ?         ?  @         @  B         B  =C         >C  kD         lD  E         E  F          Z  [          \  \          4]  *^          `^  V_          _  `         l  m         m  o          o  Lp         Np  zq         |q  r         r  s         s  u         u  2v         4v  `w         bw  x         x  y         y  z         z  |         |  F}         H}  t~         v~             Ѐ         Ҁ              ,         .  Z         \                                              @         B  n         p             ʍ         ̍             &                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               +!         :!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   A      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-y-     y--    -S;-x    S;-I-|    I-;}-\    ;}-m-    m--h    -ub-    ub-x-	    x--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               8!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 '	         (	  L
         M
  q         {                                              3          4  -          i  b                                              =  /          b  T            y              !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ,         .  Z         \                                              @         B  n	         p	  
         
                        &         (  T         V                                              :         <  h         j                                             "  N         P  |         ~                        !         !  4"         6"  b#         d#  $         $  %         %  &         &  (         (  I)         J)  w*         x*  +         +  ,         ,  .         .  //         0/  ]0         ^0  1         1  2         2  3         3  5         5  C6         D6  q7         r7  8         8  9         9  :         :  )<         *<  W=         X=  >         >  ?         ?  @         @  B         B  =C         >C  kD         lD  E         E  F         F  G         G  #I         $I  QJ         RJ  K         K  L         L  M         M  	O         
O  7P         8P  eQ         fQ  R         R  S         S  T         T  V         V  KW         LW  yX         zX  Y         Y  Z         Z  \         \  1]         2]  _^         `^  _         _  `         `  a         a  c         c  Ed         Fd  se         te  f         f  g         g  h          ~  ~          0             \  L            x                     ʐ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ;!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-:-x    :-uI-|    uI-|-\    |-%-    %--P    -b-    b-c-	    c-[-D
    [-3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               @!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         	  -         .  R         S  w	         x	  
         
           N  r         s                                                                     1            F  "          [  7          p  p                                     !          D!  D"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            *         ,  V         X                                              2         4  ^	         `	  
         
                                   :         <  f         h                                              B         D  n         p                                               J         L  v         x                          !         !  &#         (#  R$         T$  ~%         %  &         &  '         '  )         )  .*         0*  Z+         \+  ,         ,  -         -  .         .  
0         0  61         81  b2         d2  3         3  4         4  5         5  7         7  >8         @8  j9         l9  :         :  ;         ;  <         <  >         >  F?         H?  r@         t@  A         A  B         B  C         h  (j         *j  Tk         Vk  l         l  m         m  n         n  p         p  0q         2q  \r         ^r  s         s  t         t  u         u  w         w  8x         :x  dy         fy  z         z  {         {  |         |  ~         ~  @         B  l         n                      Ă                        C         E  o         q             ǉ         ɉ                                    7             V  ?          u  ^                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               <!         T!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 {
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-5-x    5-C-|    C-v-\    v-i0-    i0--h    -cb-    cb-x-	    x--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               >!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P              
          5  ?          j  t                                  	                        3  	          H	  .
          ]
  C         r                                              *                                  %         &  J         K  o                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                 L  O          ~                       %  	'         
'  9(         :(  i)         j)  *         *  +         +  ,         ,  ).         *.  Y/         Z/  0         0  1         1  2         2  4         4  I5         J5  y6         z6  7         7  8         8  	:         
:  9;         :;  i<         j<  =         =  >         >  ?         ?  )A         *A  YB         ZB  C         C  D         D  E         E  G         G  IH         JH  yI         h  i         i  *k         ,k  Zl         \l  m         m  n         n  o         o  q         q  Jr         Lr  zs         |s  t         t  u         u  
w         w  :x         <x  jy         ly  z         z  {         {  |         |  *~         ,~  Z         \                                              J         L  z         |             ڈ         ܈  
           :         <  j         P  ~                      ސ                      >         @  n         p             Ζ         Ж              .         0  ^         `                                               N         P  ~                      ޣ                      >         @  n         p             Ω         Щ              .         0  ^         `                                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               2!         <!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-:-x    :-oI-|    oI-|-\    |-q+-    q+--P    -b-    b-c-	    c-[-D
    [-3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ?!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         	  -         .  R         S  w	         x	  
         
           I  m         n                                                                     ,  	          A            V  3          k  j                                  
   	!          ?!  >"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            +         ,  W         X                                              3         4  _	         `	  
         
                                   ;         <  g         h                                              C         D  o         p                                               K         L  w         x                          !         !  '#         (#  S$         T$  %         %  &         &  '         '  )         )  /*         0*  [+         \+  ,         ,  -         -  .         .  0         0  71         81  c2         d2  3         3  4         4  5         5  7         7  ?8         @8  k9         l9  :         :  ;         ;  <         <  >         >  G?         H?  s@         t@  A         A  B         B  C         h  )j         *j  Uk         Vk  l         l  m         m  n         n  p         p  1q         2q  ]r         ^r  s         s  t         t  u         u  w         w  9x         :x  ey         fy  z         z  {         {  |         |  ~         ~  A         B  m         n             Â         Ă                        D         E  p         q             ȉ         ɉ              }                      Ҩ              ڪ                               >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               >!         T!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 {
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     --    -4-x    4-C-|    C-v-\    v-i0-    i0--h    -ib-    ib-x-	    x--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               :!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P              	          5  >          j  s                                  	                        3  	          H	  /
          ]
  D         r                                              *                                           !  E         F  j                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                  F   I!          x!  {"          "  #         %  '         
'  8(         :(  h)         j)  *         *  +         +  ,         ,  (.         *.  X/         Z/  0         0  1         1  2         2  4         4  H5         J5  x6         z6  7         7  8         8  :         
:  8;         :;  h<         j<  =         =  >         >  ?         ?  (A         *A  XB         ZB  C         C  D         D  E         E  G         G  HH         JH  xI         h  i         i  +k         ,k  [l         \l  m         m  n         n  o         o  q         q  Kr         Lr  {s         |s  t         t  u         u  w         w  ;x         <x  ky         ly  z         z  {         {  |         |  +~         ,~  [         \                                              K         L  {         |             ۈ         ܈             ;         <  k         P                        ߐ                      ?         @  o         p             ϖ         Ж              /         0  _         `                                               O         P                        ߣ                      ?         @  o         p             ϩ         Щ              /         0  _         `                                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               4!         @!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -)-    )-;-    ;-)J-    )J-w}-d    w}-G-    G--p    -b-    b-!-
    !-/-
    /--t    -9׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               =!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             $         %  I         J  n         o                       r  r                                              F  F          {  l                                                          =  a         b                          !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       <  G          x                                   %  1&         2&  ['         \'  (         (  )         )  *         *  ,         ,  --         .-  W.         X.  /         /  0         0  1         1  2          3  )4         *4  S5         T5  }6         ~6  7         7  8         8  9         9  %;         &;  O<         P<  y=         z=  >         >  ?         ?  @         @  !B         "B  KC         LC  uD         vD  E         E  F         F  G         G  I         I  FJ         HJ  pK         rK  L         L  M         M  N         N  P         P  BQ         DQ  lR         nR  S         S  T         T  U         U  W         W  >X         @X  hY         jY  Z         Z  [         [  \         \  ^         ^  :_         <_  d`         f`  a         a  b         b  c         c  e         e  6f         8f  `g         bg  h         h  i         i  j         j  l         
l  2m         4m  \n         ^n  o         o  p         p  q         q  s         s  .t         0t  Xu         Zu  v         v  w         w  x         x   z         z  *{         ,{  T|         V|  ~}         }  ~         ~                        &         (  P         R  z         |             Ά         І             "         $  L         N  v         x             ʍ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               I!         b!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    l
         m
             F      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     H
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-[-     [--    -Y;-x    Y;-I-|    I-A}-\    A}-C-    C--h    -ub-    ub--	    --h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               %!         0!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 '	         (	  L
         M
  q         {                                              3          4  .          i  c                                              =  .          b  S            x              !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            1         2  c         d                                   +         ,  ]         ^  	         	  
         
             %         &  W         X                                               Q         R                                              K         L  }         ~                                   E          F   w!         x!  "         "  #          %  &          D&  L'          '  (          (  )          )   +         G  H         H  I         I  -K         .K  _L         `L  M         M  N         N  O         O  'Q         (Q  YR         ZR  S         S  T         T  U         U  !W         "W  SX         TX  Y         Y  Z         Z  [         [  ]         ]  M^         N^  _         _  `         `  a         a  c         c  Gd         Hd  ye         ze  f         f  g         g  i         i  Aj         Bj  sk         tk  l         l  m         Pp  q         q  r         r  s         s  u         u  Hv         Jv  zw         |w  x         x  y         y  {         {  B|         D|  t}         v}  ~         ~             
           <         >  n         p             ҅         ԅ             6         8  h         j             ̋         ΋              0         2  b         d             Ƒ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               5!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -)-    )-;-    ;-)J-    )J-w}-d    w}-G-    G--p    -b-    b-!-
    !-/-
    /--t    -9׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               =!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             $         %  I         J  n         o                       r  q                                              F  E          {  m                                                         =  a         b                          !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         ,  7          h  s            	          	  
            '         %  0&         2&  Z'         \'  (         (  )         )  *         *  ,         ,  ,-         .-  V.         X.  /         /  0         0  1         1  2          3  (4         *4  R5         T5  |6         ~6  7         7  8         8  9         9  $;         &;  N<         P<  x=         z=  >         >  ?         ?  @         @   B         "B  JC         LC  tD         vD  E         E  F         F  G         G  I         I  GJ         HJ  qK         rK  L         L  M         M  N         N  P         P  CQ         DQ  mR         nR  S         S  T         T  U         U  W         W  ?X         @X  iY         jY  Z         Z  [         [  \         \  ^         ^  ;_         <_  e`         f`  a         a  b         b  c         c  e         e  7f         8f  ag         bg  h         h  i         i  j         j  	l         
l  3m         4m  ]n         ^n  o         o  p         p  q         q  s         s  /t         0t  Yu         Zu  v         v  w         w  x         x  z         z  +{         ,{  U|         V|  }         }  ~         ~                        '         (  Q         R  {         |             φ         І             #         $  M         N  w         x             ˍ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               I!         b!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    l
         m
             F      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     I
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-U-     U--    -_;-x    _;-I-|    I-G}-\    G}-C-    C--h    -ib-    ib--	    --h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               "!         .!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 '	         (	  L
         M
  q         {                                              3          4  -          i  b                                              =  /          b  T            y              !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            0         2  b         d                                   *         ,  \         ^  	         	  
         
             $         &  V         X                                               P         R                                              J         L  |         ~                                   D          F   v!         x!  "         "  #          4+  <,          p,  x-          -  .          .  /          $0  ,1         G  H         H  I         I  ,K         .K  ^L         `L  M         M  N         N  O         O  &Q         (Q  XR         ZR  S         S  T         T  U         U   W         "W  RX         TX  Y         Y  Z         Z  [         [  ]         ]  L^         N^  ~_         _  `         `  a         a  c         c  Fd         Hd  xe         ze  f         f  g         g  i         i  @j         Bj  rk         tk  l         l  m         Pp  q         q  r         r  s         s  u         u  Iv         Jv  {w         |w  x         x  y         y  {         {  C|         D|  u}         v}  ~         ~                        =         >  o         p             Ӆ         ԅ             7         8  i         j             ͋         ΋              1         2  c         d             Ǒ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               1!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     -5-    5-5-    5-D-    D-w-l    w-72-    72--x    -b-	    b-'-
    '-5-
    5--|    -ճ-    ճ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ?!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         %            J  ;          o  `                                              #  $	          X	  Y
          
                                   0         1  U         V  z                                            3         4  _                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |        $  %          2  3          3  4          5  '6          [6  j7          7  8         K  L         L  N         N  EO         FO  mP         nP  Q         Q  R         R  S         S  U         U  5V         6V  ]W         ^W  X         X  Y         Y  Z         Z  [         [  %]         &]  M^         N^  u_         v_  `         `  a         a  b         b  d         d  =e         >e  ef         ff  g         g  h         h  i         i  k         k  -l         .l  Um         Vm  }n         ~n  o         o  p         p  q         q  s         s  Et         Ft  mu         nu  v         v  w         w  x         x  z         z  5{         6{  ]|         ^|  }         }  ~         ~                        %         &  M         N  u         v             ņ         Ɔ                        =         >  e         f                        ݎ         ގ             -         .  T         V  |         ~             ̕         Ε                        D         F  l         n                                              4         6  \         ^                        Ԥ         ֤             $         &  L         N  t         v             ī         ƫ                        <         >  d         f                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               @!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    l
         m
             F      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    l
         m
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     X
            0     -,-X     ,--     -;-    ;-;-x    ;-_J-|    _J-}-\    }--    --h    -b-    b--	    --h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               F!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             $         %  I         J  n         o                                                           1  "          V  G          {                                                O  T         `                                     !          "  '#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            2         4  f         h                                   6         8  j         l  	         	  
         
             :         <  n         p                        
           >         @  r         t                                   B         D  v         x                                   F         H  z          |   !         !  "         "  $         $  %          &  3'         4'  g(         h(  )         )  *         *  ,         ,  7-         8-  k.         l.  /         /  0         0  2         2  ;3         <3  o4         p4  5         5  6         6  8         8  ?9         @9  s:         t:  ;         ;  <         <  >         >  C?         D?  w@         x@  A         A  B         B  D         D  GE         HE  {F         |F  G         G  H          lX  {Y          Y  Z          Z  \          5\  D]          x]  ^         "o  Tp         Vp  q         q  r         r  s         s  $u         &u  Xv         Zv  w         w  x         x  y         y  ({         *{  \|         ^|  }         }  ~         ~             ,         .  `         b             Ȅ         ʄ             0         2  d         f             ̊         Ί              4         6  h         j             А         Ґ             8                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               :!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     -5-    5-5-    5-D-    D-w-l    w-72-    72--x    -b-	    b-'-
    '-5-
    5--|    -ճ-    ճ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ?!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         %            J  <          o  a                                              #  %	          X	  Z
          
                                   0         1  U         V  z                                            /         0  Z                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |        $  %          8  9          $:  3;          g;  v<          <  =          =  >         K  L         L  N         N  DO         FO  lP         nP  Q         Q  R         R  S         S  U         U  4V         6V  \W         ^W  X         X  Y         Y  Z         Z  [         [  $]         &]  L^         N^  t_         v_  `         `  a         a  b         b  d         d  <e         >e  df         ff  g         g  h         h  i         i  k         k  ,l         .l  Tm         Vm  |n         ~n  o         o  p         p  q         q  s         s  Dt         Ft  lu         nu  v         v  w         w  x         x  z         z  4{         6{  \|         ^|  }         }  ~         ~                        $         &  L         N  t         v             Ć         Ɔ                        <         >  d         f                        ܎         ގ             ,         .  U         V  }         ~             ͕         Ε                        E         F  m         n                                              5         6  ]         ^                        դ         ֤             %         &  M         N  u         v             ū         ƫ                        =         >  e         f                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               >!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    l
         m
             F      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    l
         m
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     W
            0     -,-X     ,--     -A-    A-;-x    ;-_J-|    _J-}-\    }--    --h    -b-    b--	    --h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               H!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             $         %  I         J  n         o                                                           1  #          V  H          {                                                O  S         [                                     !         !  '#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            3         4  g         h                                   7         8  k         l  	         	  
         
             ;         <  o         p                                   ?         @  s         t                                   C         D  w         x                                   G         H  {          |   !         !  "         "  $         $  %          &  2'         4'  f(         h(  )         )  *         *  ,         ,  6-         8-  j.         l.  /         /  0         0  2         2  :3         <3  n4         p4  5         5  6         6  
8         8  >9         @9  r:         t:  ;         ;  <         <  >         >  B?         D?  v@         x@  A         A  B         B  D         D  FE         HE  zF         |F  G         G  H          ^  _          _  a          Aa  Pb          b  c          c  d         "o  Up         Vp  q         q  r         r  s         s  %u         &u  Yv         Zv  w         w  x         x  y         y  ){         *{  ]|         ^|  }         }  ~         ~             -         .  a         b             Ʉ         ʄ             1         2  e         f             ͊         Ί             5         6  i         j             ѐ         Ґ             9                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               8!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -S-    S-;-x    ;-5J-|    5J-}-\    }-4-    4--h    -b-    b-n-	    n--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               S!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             $         %  I         J  n         o                       r  g                                              F  ;          {  l                                                          =  a         b                          !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            $         &  J         L  p         r                                            
  .	         0	  T
         V
  z         |                                              8         :  ^         `                                                         B         D  h         j                                               &          (   L!         N!  r"          V;  X<          <  =          =  >          >  ?          .@  0A         K  L         L  M          N  %O         &O  KP         LP  qQ         rQ  R         R  S         S  T         T  	V         
V  /W         0W  UX         VX  {Y         |Y  Z         Z  [         [  \         \  ^         ^  9_         :_  _`         ``  a         a  b         b  c         c  d         d  f         f  Cg         Dg  ih         jh  i         i  j         j  k         k  m         m  'n         (n  Mo           ?         @  e         f                        י         ؙ             #         $  I         J  o         p                                              -         .  S         T  y         z             Ũ         ƨ                        7         8  ]         ^                        ϰ         а                        A         B  g                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               T!         Z!  2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     +
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -M-    M-e;-    e;-I-    I-M}-d    M}-q-    q--0    -%-    %-b-H	    b-c-P
    c-O-
    O-E׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               N!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 '	         (	  L
         M
  q         {                                              3          4  %          i  Z                                              =  .          b  S            x              !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q   w  operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       \            5         6  k         l                                   C         D  y         z  	         	  
         
             Q         R                                   )         *  _         `                                   7         8  m         n                                   E         F  {         |               !         !  #         #  S$         T$  %         %  &         &  '         '  +)         ,)  a*         b*  +         +  ,         ,  .         .  9/         :/  o0         p0  1         1  2         2  4         4  G5         H5  }6         ~6  7         7  8         8  :          :  U;         V;  <         <  =         =  >         >  -@         .@  cA         dA  B         B  C         C  E         E  ;F         <F  pG         rG  H         H  I         nn  o         o  p         p  r         r  Ds         Fs  zt         |t  u         u  v         v  x         x  Ry         Ty  z         z  {         {  |         |  *~         ,~  `         b             ́         ΁              	          :  ?          p  u                      ܇             F         H  |         ~                                           b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               R!         l!  2         "2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 o
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     - -X      -+ְ-     +ְ-Uܰ-    Uܰ-4-p    4-]@-T    ]@-h-4    h-4-    4--8    -b-    b-n-	    n--8
    -E׳-
                   =         b      J      7   U   7   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               S!         T!  a1                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 64 })  outputElements  5184    input0Elements  31104   input1Elements  24576   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     4                           #         b      J      7   M     operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       @             $         %  I         J  n         o            (	  
          ]
  Q                                              !            F  8          k  ]         d                                                                 b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     4                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  1728    input0Elements  1728    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       x            %         &  K         L  q         r                                   	         
  /	         0	  U
         V
  {         |                                              9         :  _         `                                                         C         D  i         j                                              '          (   M!         N!  s"          dA  fB          B  C          C  D          E  F         K  L         L  M          N  $O         &O  JP         LP  pQ         rQ  R         R  S         S  T         T  V         
V  .W         0W  TX         VX  zY         |Y  Z         Z  [         [  \         \  ^         ^  8_         :_  ^`         ``  a         a  b         b  c         c  d         d  f         f  Bg         Dg  hh         jh  i         i  j         j  k         k   m         m  &n         (n  Lo           >         @  d         f                        ֙         ؙ             "         $  H         J  n         p                                              ,         .  R         T  x         z             Ĩ         ƨ                        6         8  \         ^                        ΰ         а                        @         B  f                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               R!         Z!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     *
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     - -`      -%ְ-     %ְ-Oܰ-    Oܰ-i4-x    i4-'@-\    '@-h-<    h-A-    A-k-    k--    -b- 	    b-c-(
    c-O-
    O-E׳-
                   =         b      J      7   U   7   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               Q!         R!  ^1                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 64 })  outputElements  5184    input0Elements  31104   input1Elements  24576   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     4                           #         b      J      7   M     operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       @                                            '	                     !  E         F  j         k                                                /  !          d  V            {                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     4                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  1728    input0Elements  1728    invocations     1                           3         b      J      7   Q   w  operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       \            4         6  j         l                                   B         D  x         z  	         	  
         
             P         R                                   (         *  ^         `                                    6         8  l         n                                   D         F  z         |               !         !  #         #  R$         T$  %         %  &         &  '         '  *)         ,)  `*         b*  +         +  ,         ,  .         .  8/         :/  n0         p0  1         1  2         2  4         4  F5         H5  |6         ~6  7         7  8         8  :          :  T;         V;  <         <  =         =  >         >  ,@         .@  bA         dA  B         B  C         C  E         E  :F         <F  qG         rG  H         H  I         nn  o         o  p         p  r         r  Es         Fs  {t         |t  u         u  v         v  x         x  Sy         Ty  z         z  {         {  |         |  +~         ,~  a         b             ́         ΁             9         :  o         p             ۇ         ܇                        H  M          ~                                            b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     4                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               P!         h!  2         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 o
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -S-    S-5-x    5-wD-|    wD-w-\    w-7-    7--h    -b-    b-x-	    x--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               S!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         %            J  ;          o  `                                              #  	          X	  T
          
                                   0         1  U         V  z                                                       <                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                        "  
          D  ,          f  N            p         U$  %         %  &         &  '         '  ,)         -)  b*         c*  +         +  ,         ,  .         .  :/         ;/  p0         q0  1         1  2         2  4         4  H5         I5  ~6         6  7         7  8         8   :         !:  V;         W;  <         <  =         =  >         >  .@         /@  dA         eA  B         B  C         C  E         E  <F         =F  rG         G  I         I  CJ         EJ  yK         {K  L         L  M         M  O         O  QP         SP  Q         Q  R         R  S         S  )U         +U  _V         aV  W         W  X         X  Z         Z  7[         9[  m\         o\  ]         ]  ^         ^  `         `  Ea         Ga  {b         }b  c         c  d         d  f         f  Sg         Ug  h         h  i         i  j         j  +l         l  
n         ړ             E         F  {         |                                   S         T                                   +         ,  a         b             ͤ         Τ             9         :  o         p             ۪         ܪ             G         H  }         ~                                    U         V                      µ             -                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               S!         X!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 k
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -S-    S-;-x    ;-/J-|    /J-}}-\    }}--    --h    -b-    b-3n-	    3n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               P!         X!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             $         %  I         J  n         o                                                           1  "          V  G          {  }                                              O  Q         =  a         b                          !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            $         &  J         L  p         r                                            
  .	         0	  T
         V
  z         |                                              8         :  ^         `                                                         B         D  h         j                                               &          (   L!         N!  r"         t"  #         #  $         $  %         %  '         '  1(         2(  W)         X)  }*         ~*  +         +  ,         ,  -         -  /         /  ;0         <0  a1         b1  2          sG  aH          H  I          I  J          J  K          K  L         _j  k         k  l         l  m         m  n         n  p         p  Aq         Cq  gr         ir  s         s  t         t  u         u  v         w  %x         'x  Ky         My  qz         sz  {         {  |         |  }         }  	           /         1  U         W  {         }             Ǆ         Ʉ                        9         ;  _         a                        ь         ӌ                      	  .         /  T         U  z         {             ƕ         Ǖ                        8         9  ^         _                        Н         ѝ                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               Q!         b!  2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     O
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -S-    S-5-x    5-wD-|    wD-w-\    w-7-    7--h    -b-    b-x-	    x--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               S!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         %            J  <          o  a                                              #  	          X	  S
          
                                   0         1  U         V  z                                                       <                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                             	  	          2
           U$  %         %  &         &  '         '  +)         -)  a*         c*  +         +  ,         ,  .         .  9/         ;/  o0         q0  1         1  2         2  4         4  G5         I5  }6         6  7         7  8         8  :         !:  U;         W;  <         <  =         =  >         >  -@         /@  cA         eA  B         B  C         C  E         E  ;F         =F  qG         G  I         I  DJ         EJ  zK         {K  L         L  M         M  O         O  RP         SP  Q         Q  R         R  S         S  *U         +U  `V         aV  W         W  X         X  Z         Z  8[         9[  n\         o\  ]         ]  ^         ^  `         `  Fa         Ga  |b         }b  c         c  d         d  f         f  Tg         Ug  h         h  i         i  j         j  ,l         l  	n         ړ             D         F  z         |                                   R         T                                   *         ,  `         b             ̤         Τ             8         :  n         p             ڪ         ܪ             F         H  |         ~                                    T         V                      µ             ,                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               O!         T!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 k
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -S-    S-;-x    ;-/J-|    /J-}}-\    }}--    --h    -b-    b-3n-	    3n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               R!         X!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             $         %  I         J  n         o                                                           1  #          V  H          {  |                                              O  P         =  a         b                          !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            %         &  K         L  q         r                                   	         
  /	         0	  U
         V
  {         |                                              9         :  _         `                                                         C         D  i         j                                              '          (   M!         N!  s"         t"  #         #  $         $  %         %  
'         '  0(         2(  V)         X)  |*         ~*  +         +  ,         ,  -         -  /         /  :0         <0  `1         b1  2          M  N          ?N  -O          aO  OP          P  qQ          Q  R         _j  k         k  l         l  m         m  n         n  p         p  Bq         Cq  hr         ir  s         s  t         t  u         u   w         w  &x         'x  Ly         My  rz         sz  {         {  |         |  }         }  
           0         1  V         W  |         }             Ȅ         Ʉ                        :         ;  `         a                        Ҍ         ӌ                      	  -         /  S         U  y         {             ŕ         Ǖ                        7         9  ]         _                        ϝ         ѝ                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               U!         f!  2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     N
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     ---X     ---     -Y-    Y-:-x    :-I-|    I-|-\    |-;-    ;--h    -]b-    ]b---	    ---h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               U!         \!  2         
2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         	  -         .  R         S  w	         x	  
         
           N  r         s                                                                      1            F  *          [  ?          p  x                                     !          D!  L"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            2         4  f         h                                   6         8  j         l  	         	  
         
             :         <  n         p                        
           >         @  r         t                                   B         D  v         x                                   F         H  z          |   !         !  "         "  $         $  J%         L%  ~&         &  '         '  (         (  *         *  N+         P+  ,         ,  -         -  .         .  0          0  R1         T1  2         2  3         3  4         4  "6         $6  V7         X7  8         8  9         9  :         :  &<         (<  Z=         \=  >         >  ?         ?  @         @  *B         ,B  ^C         `C  D         D  E         E  F         F  .H         0H  bI         dI  J         J  K         K  L          M  2N         4N  fO         hO  P         P  Q         Q  S         S  6T         8T  jU         lU  V         W  QX         SX  Y         Y  Z         Z  [         [  !]         #]  U^         W^  _         _  `         `  a         a  %c         'c  Yd         [d  e         e  f         f  g         g  )i         +i  ]j          x  *y          [y  mz          z  {          {  |          $}  6~         T                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ,!         8!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   J      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -;-    ;-5-x    5-C-|    C-v-\    v-I-    I--h    -b-    b--n-	    -n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               H!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        5  7          j  l                                  	                        3  	          H	  &
          ]
  ;         r                                              *                                  %         &  J         K  o                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            &         (  N         P  v         x                                              >	         @	  f
         h
                                              .         0  V         X  ~                                                        F         H  n         p                                              6         8  ^          `   !         !  "         "  $         $  :%         <%  b&         d&  '         '  (         (  )         )  +         +  *,         ,,  R-         T-  z.         |.  /         /  0         0  1         1  3         3  B4         D4  j5         l5  6         6  7         7  8         8  
:         :  2;         4;  Z<         \<  =         =  >         >  ?         ?  @         @  "B         $B  JC         LC  rD         tD  E         E  F          R  S          S  T          U  V          NV  LW          W  ~X         Bm  hn         jn  o         o  p         p  q         q  s         
s  0t         2t  Xu         Zu  v         v  w         w  x         x  y         y   {         "{  H|         J|  p}         r}  ~         ~                                   8         :  `         b                        ؇         ڇ              (         *  P         R  x         z             Ȏ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               F!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     O
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -M-    M-:-x    :-{I-|    {I-|-\    |-;-    ;--h    -cb-    cb-'-	    '--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               Q!         X!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         	  -         .  R         S  w	         x	  
         
           I  m         n                                                                     ,            A  &          V  ;          k  r                                  
   !          ?!  F"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            3         4  g         h                                   7         8  k         l  	         	  
         
             ;         <  o         p                                   ?         @  s         t                                   C         D  w         x                                   G         H  {          |   !         !  "         "  $         $  K%         L%  &         &  '         '  (         (  *         *  O+         P+  ,         ,  -         -  .         .  0          0  S1         T1  2         2  3         3  4         4  #6         $6  W7         X7  8         8  9         9  :         :  '<         (<  [=         \=  >         >  ?         ?  @         @  +B         ,B  _C         `C  D         D  E         E  F         F  /H         0H  cI         dI  J         J  K         K  L          M  3N         4N  gO         hO  P         P  Q         Q  S         S  7T         8T  kU         lU  V         W  RX         SX  Y         Y  Z         Z  [         [  "]         #]  V^         W^  _         _  `         `  a         a  &c         'c  Zd         [d  e         e  f         f  g         g  *i         +i  ^j          g~  y                                  0  B          s           T                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               .!         <!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   I      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -A-    A-5-x    5-C-|    C-v-\    v-I-    I--h    -b-    b--n-	    -n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               J!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        5  6          j  k                                  	                        3  	          H	  '
          ]
  <         r                                              *                                           !  E         F  j                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            '         (  O         P  w         x                                              ?	         @	  g
         h
                                              /         0  W         X                                                          G         H  o         p                                              7         8  _          `   !         !  "         "  $         $  ;%         <%  c&         d&  '         '  (         (  )         )  +         +  +,         ,,  S-         T-  {.         |.  /         /  0         0  1         1  3         3  C4         D4  k5         l5  6         6  7         7  8         8  :         :  3;         4;  [<         \<  =         =  >         >  ?         ?  @         @  #B         $B  KC         LC  sD         tD  E         E  F          X  Y          Y  Z          [  \          H\  F]          z]  x^         Bm  in         jn  o         o  p         p  q         q  	s         
s  1t         2t  Yu         Zu  v         v  w         w  x         x  y         y  !{         "{  I|         J|  q}         r}  ~         ~                                   9         :  a         b                        ه         ڇ             )         *  Q         R  y         z             Ɏ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               H!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     N
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-;-x    ;-)J-|    )J-w}-\    w}--    --h    -Kb-    Kb-'-	    '--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               G!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             $         %  I         J  n         o                       r  z                                              F  N          {  l                                                          =  a         b                          !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            0         2  b         d                                   *         ,  \         ^  	         	  
         
             $         &  V         X                                               P         R                                              J         L  |         ~                                   D          F   v!         x!  "         "  #          q<  }=          =  >          >  ?          @  +A          YA  eB         F  G         G  &I         (I  XJ         ZJ  K         K  L         L  M         M   O         "O  RP         TP  Q         Q  R         R  S         S  U         U  LV         NV  ~W         W  X         X  Y         Y  [         [  F\         H\  x]         z]  ^         ^  _         _  a         a  @b         Bb  rc         tc  d         d  e         e  g         
g  :h         <h  li         ni  j         j  k         k  m         m  5n         6n  go         ho  p         p  q         q  r         r  /t         0t  au         bu  v         v  w         w  x         x  )z         *z  [{         \{  |         |  }         }  ~         ~  #         $  U         V                                              O         P                                              I         J  {                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               '!         .!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   J      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -;-    ;-Y;-x    Y;-I-|    I-A}-\    A}-o-    o--h    -b-    b-x-	    x--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               F!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 '	         (	  L
         M
  q         {                                              3          4  6          i  k                                    
          =  .          b  S            x              !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            )         *  S         T  }         ~                                   %         &  O	         P	  y
         z
                                   !         "  K         L  u         v                                              G         H  q         r                                              C         D  m         n               !         !  "         $  <%         >%  f&         h&  '         '  (         (  )         )  +         +  8,         :,  b-         d-  .         .  /         /  0         0  
2         2  43         63  ^4         `4  5         5  6         6  7         7  9         9  0:         2:  Z;         \;  <         <  =         =  >         >  @         @  ,A         .A  VB         XB  C         C  D         D  E         E  F          G  (H         :H  bI         dI  J         J  K         K  L         L  
N         N  4O         6O  ^P         `P  Q         Q  R         R  S         S  U         U  0V         2V  ZW         \W  X         X  Y         Y  Z         Z  \         \  ,]         .]  V^         X^  _         _  `         `  a         a  b          c  (d         *d  Re         Te  |f         ~f  g         g  h            Ȃ                      6  <          p  v                     ʎ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               C!         V!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-;-x    ;-)J-|    )J-w}-\    w}--    --h    -Qb-    Qb-'-	    '--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               C!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             $         %  I         J  n         o                       r  y                                              F  M          {  m                                                         =  a         b                          !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            1         2  c         d                                   +         ,  ]         ^  	         	  
         
             %         &  W         X                                               Q         R                                              K         L  }         ~                                   E          F   w!         x!  "         "  #          B  C          C  D          E  F         F  G         G  'I         (I  YJ         ZJ  K         K  L         L  M         M  !O         "O  SP         TP  Q         Q  R         R  S         S  U         U  MV         NV  W         W  X         X  Y         Y  [         [  G\         H\  y]         z]  ^         ^  _         _  a         a  Ab         Bb  sc         tc  d         d  e         e  	g         
g  ;h         <h  mi         ni  j         j  k         k  m         m  4n         6n  fo         ho  p         p  q         q  r         r  .t         0t  `u         bu  v         v  w         w  x         x  (z         *z  Z{         \{  |         |  }         }  ~         ~  "         $  T         V                                              N         P                                              H         J  z          ?  K          y                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               )!         2!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   I      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-_;-x    _;-I-|    I-G}-\    G}-o-    o--h    -b-    b-x-	    x--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               D!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 '	         (	  L
         M
  q         {                                              3          4  5          i  j                                    	          =  /          b  T            y              !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            (         *  R         T  |         ~                                   $         &  N	         P	  x
         z
                                             "  J         L  t         v                                              F         H  p         r                                              B         D  l         n               !         !  "         $  =%         >%  g&         h&  '         '  (         (  )         )  +         +  9,         :,  c-         d-  .         .  /         /  0         0  2         2  53         63  _4         `4  5         5  6         6  7         7  9         9  1:         2:  [;         \;  <         <  =         =  >         >  @         @  -A         .A  WB         XB  C         C  D         D  E         E  F          G  )H         :H  cI         dI  J         J  K         K  L         L  N         N  5O         6O  _P         `P  Q         Q  R         R  S         S  U         U  1V         2V  [W         \W  X         X  Y         Y  Z         Z  \         \  -]         .]  W^         X^  _         _  `         `  a         a  b          c  )d         *d  Se         Te  }f         ~f  g         g  h                        $          X  ^                      ̌  ҍ         ʎ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               G!         Z!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-:-x    :-{I-|    {I-|-\    |-
-    
--P    -ub-    ub-c-	    c-g-D
    g-3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               >!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         	  -         .  R         S  w	         x	  
         
           N  r         s                                                                      1            F  *          [  ?          p  x                                     !          D!  L"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            .         0  ^         `                                               N         P  ~	         	  
         
                        >         @  n         p                                    .         0  ^         `                                               N         P  ~                                              >!         @!  n"          p"  ]#          #  {$          $  %         %  &         &  ((         *(  X)         Z)  *         *  +         +  ,         ,  .         .  H/         J/  x0         z0  1         1  2         2  4         
4  85         :5  h6         j6  7         7  8         8  9         9  (;         *;  X<         Z<  =         =  >         >  ?         ?  A         A  HB         JB  xC         zC  D         D  E         E  G         
G  8H          :H  'I          XI  EJ         h   j         j  0k         2k  `l         bl  m         m  n         n  o         o   q         "q  Pr         Rr  s         s  t         t  u         u  w         w  @x         Bx  py         ry  z         z  {         {   }         }  0~         2~  `         b                                              L         N  {         }             ۈ         ݈             ;                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               6!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 {
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-5-x    5-C-|    C-v-\    v--    --h    -{b-    {b-x-	    x--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               F!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        5  7          j  l                                  	                        3  	          H	  &
          ]
  ;         r                                              *                                  %         &  J         K  o                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                        .  %          \  S                                 &  '         '  (         (  
*         *  6+         7+  b,         c,  -         -  .         .  /         /  1         1  >2         ?2  j3         k3  4         4  5         5  6         6  8         8  F9         G9  r:         s:  ;         ;  <         <  =         =  !?         #?  M@         O@  yA         {A  B         B  C         C  D         D  )F         +F  UG         WG  H         H  I         I  J         J  L         L  1M         3M  ]N         _N  O         O  P         P  Q         Q  S         S  9T         ;T  eU         gU  V         V  W         W  X         X  Z         Z  A[         C[  m\         o\  ]         ]  ^         ^  _         _  a         a  Ib         Kb  uc         wc  d         d  e         e  f         f  %h         'h  Qi         Si  }j         j  k         k  l          n  *o         ,o  Vp         Xp  q         q  r         r  s         s  u         u  2v         4v  ^w         `w  x         x  y         y  z         z  |         |  :}         <}  f~         h~                                              B         D  n         p             Ƈ         ȇ                         J         L  v         x             Ύ         Ў             &                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               :!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     --    -:-x    :-uI-|    uI-|-\    |-
-    
--P    -ub-    ub-c-	    c-g-D
    g-3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ;!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         	  -         .  R         S  w	         x	  
         
           I  m         n                                                                     ,            A  &          V  ;          k  r                                  
   !          ?!  F"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            /         0  _         `                                               O         P  	         	  
         
                        ?         @  o         p                                    /         0  _         `                                               O         P                                                ?!         @!  o"         %  &         &  )(         *(  Y)         Z)  *         *  +         +  ,         ,  .         .  I/         J/  y0         z0  1         1  2         2  	4         
4  95         :5  i6         j6  7         7  8         8  9         9  );         *;  Y<         Z<  =         =  >         >  ?         ?  A         A  IB         JB  yC         zC  D         D  E         E  	G         
G  9H          vJ  cK          K  L          L  M          M  N          N  O         h  j         j  1k         2k  al         bl  m         m  n         n  o         o  !q         "q  Qr         Rr  s         s  t         t  u         u  w         w  Ax         Bx  qy         ry  z         z  {         {  }         }  1~         2~  a         b                                              M         N  |         }             ܈         ݈             <                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               6!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 {
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-4-x    4-C-|    C-v-\    v--    --h    -{b-    {b-x-	    x--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               D!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        5  6          j  k                                  	                        3  	          H	  '
          ]
  <         r                                              *                                           !  E         F  j                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                 B  9	          p	  g
          
           &  '         '  (         (  	*         *  5+         7+  a,         c,  -         -  .         .  /         /  1         1  =2         ?2  i3         k3  4         4  5         5  6         6  8         8  E9         G9  q:         s:  ;         ;  <         <  =         =  "?         #?  N@         O@  zA         {A  B         B  C         C  D         D  *F         +F  VG         WG  H         H  I         I  J         J  L         L  2M         3M  ^N         _N  O         O  P         P  Q         Q  S         S  :T         ;T  fU         gU  V         V  W         W  X         X  Z         Z  B[         C[  n\         o\  ]         ]  ^         ^  _         _  a         a  Jb         Kb  vc         wc  d         d  e         e  f         f  &h         'h  Ri         Si  ~j         j  k         k  l          n  +o         ,o  Wp         Xp  q         q  r         r  s         s  u         u  3v         4v  _w         `w  x         x  y         y  z         z  |         |  ;}         <}  g~         h~                                              C         D  o         p             Ǉ         ȇ                         K         L  w         x             ώ         Ў             '                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               <!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,-y-     y--    -;-    ;-#J-    #J-q}-l    q}-!-    !--P    --    -b-h	    b--p
    -)-
    )--    -3׳-L                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               9!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             $         %  I         J  n         o                       r  z                                              F  N          {  l                                                          =  a         b                          !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        %  &         &  '         '  )         
)  6*         8*  d+         f+  ,         ,  -         -  .         .  0         0  J1         L1  x2         z2  3         3  4         4  6         6  07         27  ^8         `8  9         9  :         :  ;         ;  =         =  D>         F>  r?         t?  @         @  A         A  B         B  *D         ,D  XE         ZE  F         F  G         G  H         H  J         J  >K         @K  lL         nL  M         M  N         N  O         O  %Q         &Q  SR         TR  S         S  T         T  U         U  W         W  9X         :X  gY         hY  Z         Z  [         [  \         \  ^          ^  M_         N_  {`         |`  a         a  b         b  d         d  3e         4e  af         bf  g         g  h         h  i         i  k         k  Gl         Hl  um         vm  n         n  o         o  p          q  ,r         .r  Zs         \s  t         t  u         u  v         v  x          x  y          By  ?z          pz  m{          {  |          |  }         }  &         (  T         V                        ރ                      :         <  h         j             ĉ         Ɖ                       "  N         P  |         ~                                b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               C!         Z!  1         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    l
         m
             F                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     P
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-s-     s--    -S;-x    S;-I-|    I-;}-\    ;}-,-    ,--h    -{b-    {b-傳-	    傳--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               4!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 '	         (	  L
         M
  q         {                                              3          4  6          i  k                                    
          =  .          b  S            x              !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            -         .  [         \                                              A         B  o	         p	  
         
                        '         (  U         V                                              ;         <  i         j                                   !         "  O         P  }         ~                        !         !  5"         6"  c#         %  &         &  '         '  %)         &)  S*         T*  +         +  ,         ,  -         -  /         /  90         :0  g1         h1  2         2  3         3  4         4  6          6  M7         N7  {8         |8  9         9  :         :  <         <  3=         4=  a>         b>  ?         ?  @         @  A         A  C         C  GD         HD  uE         vE  F         F  G         G  H         TI  J         J  K          :X  4Y          hY  bZ          Z  [          [  \          \  ]           ?         A  m         o             ɐ         ː             %         '  S         U                        ݗ         ߗ             9         ;  g         i             Ý         ŝ                      !  M         O  {         }             פ         ٤             3         5  a         c                                                      >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               =!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,-s-     s--    -;-    ;-#J-    #J-q}-l    q}-!-    !--P    --    -b-h	    b--p
    -)-
    )--    -3׳-L                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               7!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             $         %  I         J  n         o                       r  y                                              F  M          {  m                                                         =  a         b                          !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        %  &         &  '         '  	)         
)  7*         8*  e+         f+  ,         ,  -         -  .         .  0         0  K1         L1  y2         z2  3         3  4         4  6         6  17         27  _8         `8  9         9  :         :  ;         ;  =         =  E>         F>  s?         t?  @         @  A         A  B         B  +D         ,D  YE         ZE  F         F  G         G  H         H  J         J  ?K         @K  mL         nL  M         M  N         N  O         O  $Q         &Q  RR         TR  S         S  T         T  U         U  
W         W  8X         :X  fY         hY  Z         Z  [         [  \         \  ^          ^  L_         N_  z`         |`  a         a  b         b  d         d  2e         4e  `f         bf  g         g  h         h  i         i  k         k  Fl         Hl  tm         vm  n         n  o         o  p          q  -r         .r  [s         \s  t         t  u         u  v         v  x         x  Ay         By  oz         pz  {         {  |         |  }          }  ~          (  %          V  S                                              ;         <  i         j             ŉ         Ɖ             !         "  O         P  }         ~                                b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               C!         Z!  1         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    l
         m
             F                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     Q
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-y-     y--    -S;-x    S;-I-|    I-;}-\    ;}-,-    ,--h    -ub-    ub-낳-	    낳--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               8!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 '	         (	  L
         M
  q         {                                              3          4  5          i  j                                    	          =  /          b  T            y              !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ,         .  Z         \                                              @         B  n	         p	  
         
                        &         (  T         V                                              :         <  h         j                                             "  N         P  |         ~                        !         !  4"         6"  b#         %  &         &  '         '  $)         &)  R*         T*  +         +  ,         ,  -         -  
/         /  80         :0  f1         h1  2         2  3         3  4         4  6          6  L7         N7  z8         |8  9         9  :         :  <         <  2=         4=  `>         b>  ?         ?  @         @  A         A  C         C  FD         HD  tE         vE  F         F  G         G  H         TI  J         J  K           ^  _          N_  H`          |`  va          a  b          b  c           @         A  n         o             ʐ         ː             &         '  T         U                        ޗ         ߗ             :         ;  h         i             ĝ         ŝ                       !  N         O  |         }             ؤ         ٤             4         5  b         c                                                      >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               9!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,-s-     s--    -5-    5-eD-    eD-w-l    w-4-    4--x    -ib-	    ib--
    -#-
    #-s-|    s-ճ-    ճ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               7!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         %            J  ;          o  `                                              #  	          X	  T
          
                                   0         1  U         V  z                                                       <                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       #          T  a                                 D%  n&         JL  uM         vM  N         N  O         O  P         P  %R         &R  QS         RS  }T         ~T  U         U  V         V  X         X  -Y         .Y  YZ         ZZ  [         [  \         \  ]         ]  	_         
_  5`         6`  aa         ba  b         b  c         c  d         d  f         f  =g         >g  ih         jh  i         i  j         j  k         k  m         m  En         Fn  qo         ro  p         p  q         q  r         r  !t         "t  Mu         Nu  yv         zv  w         w  x         x  y         y  ){         *{  U|         V|  }         }  ~         ~                        1         2  ]         ^                                              9         :  e         f                                              A         B  m         n             ē         Ɠ                        H         J  t         v             ̚         Κ             $         &  P         R  |         ~             ԡ         ֡              ,         .  X         Z                        ܨ         ި           
  4         6  `         b                                              <         >  h         j                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               8!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    l
         m
             F      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    l
         m
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            0     -,-X     ,--     -)-    )-;-x    ;-J-|    J-k}-\    k}-{-    {--h    -]b-    ]b-傳-	    傳--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               B!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             $         %  I         J  n         o                                                           1  "          V  G          {  }                                              O  Q         =  a         b                          !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            .         0  ^         `                                               N         P  ~	         	  
         
                        >         @  n         p                                    .         0  ^         `                                               N         P  ~                                              >!         @!  n"         p"  #         D%  s&         t&  '         '  (         (  *         *  3+         4+  c,         d,  -         -  .         .  /         /  #1         $1  S2         T2  3         3  4         4  5         5  7         7  C8         D8  s9         t9  :         :  ;         ;  =         =  3>         4>  c?         d?  @         @  A         A  B         B  #D         $D  SE         TE  F         F  G         G  H          a  b          b  d          8d  Be          ve  f          f  g         ir  s         s  t         t  u         u  'w         )w  Wx         Yx  y         y  z         z  {         {  }         }  G~         I~  w         y             ׁ         ف           	  7         9  g         i             Ǉ         ɇ             '         )  W         Y                                              G         I  w         y             ה         ٔ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               4!         >!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,-s-     s--    -5-    5-eD-    eD-w-l    w-4-    4--x    -cb-	    cb--
    -#-
    #-s-|    s-ճ-    ճ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               5!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         %            J  <          o  a                                              #  	          X	  S
          
                                   0         1  U         V  z                                                       <                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                      L   Y!          !  "          "  #          $  %         D%  o&         JL  tM         vM  N         N  O         O  P         P  $R         &R  PS         RS  |T         ~T  U         U  V         V   X         X  ,Y         .Y  XZ         ZZ  [         [  \         \  ]         ]  _         
_  4`         6`  `a         ba  b         b  c         c  d         d  f         f  <g         >g  hh         jh  i         i  j         j  k         k  m         m  Dn         Fn  po         ro  p         p  q         q  r         r   t         "t  Lu         Nu  xv         zv  w         w  x         x  y         y  ({         *{  T|         V|  }         }  ~         ~                        0         2  \         ^                                              8         :  d         f                                              @         B  l         n             œ         Ɠ                        I         J  u         v             ͚         Κ             %         &  Q         R  }         ~             ա         ֡             -         .  Y         Z                        ݨ         ި  	         
  5         6  a         b                                              =         >  i         j                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               6!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    l
         m
             F      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    l
         m
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            0     -,-X     ,--     -)-    )-;-x    ;-J-|    J-k}-\    k}-{-    {--h    -]b-    ]b-傳-	    傳--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               B!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             $         %  I         J  n         o                                                           1  #          V  H          {  |                                              O  P         =  a         b                          !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            /         0  _         `                                               O         P  	         	  
         
                        ?         @  o         p                                    /         0  _         `                                               O         P                                                ?!         @!  o"         p"  #         D%  r&         t&  '         '  (         (  *         *  2+         4+  b,         d,  -         -  .         .  /         /  "1         $1  R2         T2  3         3  4         4  5         5  7         7  B8         D8  r9         t9  :         :  ;         ;  =         =  2>         4>  b?         d?  @         @  A         A  B         B  "D         $D  RE         TE  F         F  G         G  H          g  h          0i  :j          nj  xk          k  l          l  m         ir  s         s  t         t  u         u  (w         )w  Xx         Yx  y         y  z         z  {         {  }         }  H~         I~  x         y             ؁         ف           	  8         9  h         i             ȇ         ɇ             (         )  X         Y                                              H         I  x         y             ؔ         ٔ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               2!         >!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-y-     y--    -};-x    };-J-|    J-e}-\    e}-96-    96--h    -b-    b-m-	    m--h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               8!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             $         %  I         J  n         o                       r  z                                              F  N          {  l                                                          =  a         b                          !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            (         *  R         T  |         ~                                   $         &  N	         P	  x
         z
                                             "  J         L  t         v                                              F         H  p         r                                              B         D  l         n               !         !  "          #  $          %  &          @&  >'          r'  p(          (  )         K  L         L  N         N  1O         2O  [P         \P  Q         Q  R         R  S         S  U         U  -V         .V  WW         XW  X         X  Y         Y  Z         Z  [          \  )]         *]  S^         T^  }_         ~_  `         `  a         a  b         b  %d         &d  Oe         Pe  yf         zf  g         g  h         h  i         i  !k         "k  Kl         Ll  um         vm  n         n  o           6         8  `         b                        ޚ                    
  2         4  \         ^                        ڡ         ܡ             .         0  X         Z                        ֨         ب              *         ,  T         V  ~                      ү         ԯ             &         (  P         R  z         |             ζ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               Q!         R!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     +
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -;-    ;-M;-    M;-I-    I-5}-d    5}--    -U-0    U--    -b-H	    b-c-P
    c-=-
    =-'׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               S!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 '	         (	  L
         M
  q         {                                              3          4  6          i  k                                    
          =  .          b  S            x              !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q   w  operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       \            1         2  c         d                                   +         ,  ]         ^  	         	  
         
             %         &  W         X                                               Q         R                                              K         L  }         ~                                   E          F   w!         x!  "          "  #         #  %         %  ?&         @&  q'         r'  (         (  )         )  +         +  9,         :,  k-         l-  .         .  /         /  1         1  32         42  e3         f3  4         4  5         5  6         6  -8         .8  _9         `9  :         :  ;         ;  <         <  '>         (>  Y?         Z?  @         @  A         A  B         B  !D         "D  SE         TE  F          F  G          G  H          H  I          ir  js         s  t         t  u         u  /w         1w  ax         cx  y         y  z         z  {         {  )}         +}  [~         ]~                                   #         %  U         W                                              O         Q                                              I                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               [!         \!  	2         
2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 o
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-y-     y--    -};-x    };-J-|    J-e}-\    e}-?6-    ?6--h    -b-    b-m-	    m--h
    -'׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               :!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             $         %  I         J  n         o                       r  y                                              F  M          {  m                                                         =  a         b                          !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            )         *  S         T  }         ~                                   %         &  O	         P	  y
         z
                                   !         "  K         L  u         v                                              G         H  q         r                                              C         D  m         n               !         !  "          )  *          +  ,          :,  8-          l-  j.          .  /         K  L         L  N         N  0O         2O  ZP         \P  Q         Q  R         R  S         S  U         U  ,V         .V  VW         XW  X         X  Y         Y  Z         Z  [          \  (]         *]  R^         T^  |_         ~_  `         `  a         a  b         b  $d         &d  Ne         Pe  xf         zf  g         g  h         h  i         i   k         "k  Jl         Ll  tm         vm  n         n  o           7         8  a         b                        ߚ           	         
  3         4  ]         ^                        ۡ         ܡ             /         0  Y         Z                        ר         ب             +         ,  U         V                        ӯ         ԯ             '         (  Q         R  {         |             ϶                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               Q!         R!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     *
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -5-    5-M;-    M;-I-    I-5}-d    5}--    -U-0    U--    -b-H	    b-c-P
    c-=-
    =-'׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               P!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 '	         (	  L
         M
  q         {                                              3          4  5          i  j                                    	          =  /          b  T            y              !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q   w  operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       \            0         2  b         d                                   *         ,  \         ^  	         	  
         
             $         &  V         X                                               P         R                                              J         L  |         ~                                   D          F   v!         x!  "         "  #         #  %         %  >&         @&  p'         r'  (         (  )         )  +         +  8,         :,  j-         l-  .         .  /         /   1         1  22         42  d3         f3  4         4  5         5  6         6  ,8         .8  ^9         `9  :         :  ;         ;  <         <  &>         (>  X?         Z?  @         @  A         A  B         B   D         "D  RE         TE  F         F  G         G  H         H  J         ir  s          s  t          t  u          u   w          1w  2x          cx  dy         y  z         z  {         {  *}         +}  \~         ]~                                   $         %  V         W                                              P         Q                                              J                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               W!         X!  2         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 o
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-C-     C--    -5-x    5-wD-|    wD-w-\    w-+-    +--h    -b-    b-m-	    m--h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     `            !          P!  1          1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         #            F  3          i  V            y                                    !	          T	  X
          
                                   6         7  ]         ^                                              3         4  Z                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            4         6  j         l                                   B         D  x         z  	         	  
         
             P         R                                   (         *  ^         `                                    6         8  l         n                                   D         F  z         |               !         !  #         #  R$          0  1          1  2          03  24          h4  j5          5  6          I  UJ         VJ  K         K  L         L  M         M  -O         .O  cP         dP  Q         Q  R         R  T         T  ;U         <U  qV         rV  W         W  X         X  Z         Z  I[         J[  \         \  ]         ]  ^         ^  !`         "`  Wa         Xa  b         b  c         c  d         d  /f         0f  eg         fg  h         h  i         i  k         k  =l         >l  sm         tm  n         n  o         o  q         q  Kr         Lr  s         s  t         t  u         u  #w         $w  Yx         Zx  y         y  z         z  {         {  1}         2}  g~         h~             Ӏ         Ԁ  	         
  ?         @  u         v                                   M         N                                   %         &  [         \             Ǒ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               O!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     +
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,-C-     C--    -e;-x    e;-I-|    I-M}-\    M}--    --h    -ub-    ub-ӂ-	    ӂ--h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     g            !          J!  1          1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             &         '  M         N  t         u                                                           +            N  ;          q  w                                  
            =  C           E         F  l         m               !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            6         8  n         p                                   N         P             	         	  
         
  .         0  f         h                                   F         H  ~                                 &         (  ^         `                                   >         @  v         x                          "          "  V#         X#  $         $  %         %  &          '  7(         8(  o)         p)  *         *  +         +  -         -  O.         P.  /         /  0         0  1         1  /3         03  g4         h4  5         5  6         6  8         8  G9         H9  :         :  ;         ;  <         <  '>         (>  _?         `?  @         @  A         A  C         C  ?D         @D  wE         xE  F         F  G         G  I          e  f          f  h          8h  Ei          {i  j          j  k         o   q         q  8r         :r  ps         rs  t         t  u         u  w         w  Px         Rx  y         y  z         z  {         {  0}         2}  h~         j~             ؀         ڀ             H         J                                   (         *  `         b             Ћ         ҋ           
  @         B  x         z                                  "  X                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               I!         J!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,-C-     C--    -5-x    5-wD-|    wD-w-\    w-+-    +--h    -b-    b-m-	    m--h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     a            	!          P!  1          1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         #            F  4          i  W            z                                     	          T	  W
          
                                   6         7  ]         ^                                              3         4  Z                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            5         6  k         l                                   C         D  y         z  	         	  
         
             Q         R                                   )         *  _         `                                   7         8  m         n                                   E         F  {         |               !         !  #         #  S$          6  7          8  9          H9  J:          :  ;          ;  <          I  TJ         VJ  K         K  L         L  M         M  ,O         .O  bP         dP  Q         Q  R         R  T         T  :U         <U  pV         rV  W         W  X         X  Z         Z  H[         J[  ~\         \  ]         ]  ^         ^   `         "`  Va         Xa  b         b  c         c  d         d  .f         0f  dg         fg  h         h  i         i  k         k  <l         >l  rm         tm  n         n  o         o  q         q  Jr         Lr  s         s  t         t  u         u  "w         $w  Xx         Zx  y         y  z         z  {         {  0}         2}  f~         h~             Ҁ         Ԁ           
  >         @  t         v                                   L         N                                   $         &  Z         \             Ƒ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               O!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     *
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,-I-     I--    -e;-x    e;-I-|    I-M}-\    M}--    --h    -ub-    ub-ӂ-	    ӂ--h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     h            !          N!  1          1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             &         '  M         N  t         u                                                           +            N  <          q  v                                  
            =  B           E         F  l         m               !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            7         8  o         p                                   O         P             	         	  
         
  /         0  g         h                                   G         H                                   '         (  _         `                                   ?         @  w         x                          "          "  W#         X#  $         $  %         %  &          '  6(         8(  n)         p)  *         *  +         +  -         -  N.         P.  /         /  0         0  1         1  .3         03  f4         h4  5         5  6         6  8         8  F9         H9  ~:         :  ;         ;  <         <  &>         (>  ^?         `?  @         @  A         A  C         C  >D         @D  vE         xE  F         F  G         G  I          l  m          Dm  Qn          n  o         o  q         q  9r         :r  qs         rs  t         t  u         u  w         w  Qx         Rx  y         y  z         z  {         {  1}         2}  i~         j~             ـ         ڀ             I         J                                   )         *  a         b             ы         ҋ  	         
  A         B  y         z                        !         "  Y          a  n                               >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               I!         J!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,-O-     O--    -:-x    :-cI-|    cI-|-\    |-_
-    _
--h    -b-    b-ex-	    ex--h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     b            !          T!  1          1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P           %         &  L         M  s	         t	  
         
           D  j         k                                                                     5            L  (          c  ?          z  v                                     !          F!  B"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            2         4  f         h                                   6         8  j         l  	         	  
         
             :         <  n         p                        
           >         @  r         t                                   B         D  v         x                                   F         H  z          |   !         !  "         "  $         $  J%         L%  ~&         &  '         '  (         (  *         *  N+         P+  ,         ,  -         -  .         .  0          0  R1         T1  2         2  3         3  4         4  "6         $6  V7         X7  8         8  9         9  :         :  &<         (<  Z=         \=  >         >  ?         ?  @         @  *B         ,B  ^C         `C  D         D  E         E  F         F  .H         ^H  I         I  J         J  K         K  -M         .M  aN         bN  O         O  P         P  Q         Q  1S         2S  eT         fT  U         U  V         V  X         X  5Y         6Y  iZ         jZ  [         [  \         \  ^         ^  9_         :_  m`         n`  a         a  b         b  	d         
d  =e         >e  qf         rf  g         g  h         h  j         j  Ak         Bk  ul          ~  g                      Ƈ              Ӊ                               >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               R!         V!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 i
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            <     -,-p     ,-y-     y--    -4-    4-C-    C-v-t    v---     --C5-X    C5--    -Eb-p	    Eb--x
    -#-
    #-s-    s-ճ-\    ճ--<                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               =!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        3  9          f  l                                                          -  	          D	  *
          [
  A         r                                              4                                 +         ,  R         S  y                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h         V"  J#          }#  q$          H  tI         I  J         J  K         K  M         M  EN         GN  mO         oO  P         P  Q         Q  R         R  T         T  5U         7U  ]V         _V  W         W  X         X  Y         Y  Z         Z  %\         '\  M]         O]  u^         w^  _         _  `         `  a         a  c         c  =d         ?d  ee         ge  f         f  g         g  h         h  j         j  -k         /k  Ul         Wl  }m         m  n         n  o         o  p         p  r         r  Es         Gs  mt         ot  u         u  v         v  w         w  y         y  5z         7z  ]{         _{  |         |  }         }  ~         ~             %         '  M         O  u         w             Ņ         ǅ                        =         ?  e         g                        ݍ         ߍ                        .  "         U  {         }             ˔         ͔                        C         E  k         m                                              3         5  [         ]                        ӣ         գ             #         %  K         M  s         u             ê         Ū                        ;                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     g            !          P!  1          1  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    n
         o
             L                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    p
         q
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            0     -,-X     ,-I-     I--    -:-x    :-]I-|    ]I-|-\    |--    --h    -b-    b-ex-	    ex--h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     a            	!          P!  1          1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P           %         &  L         M  s	         t	  
         
           ?  e         f                                                                     0            G  $          ^  ;          u  p                                     	!          A!  <"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            3         4  g         h                                   7         8  k         l  	         	  
         
             ;         <  o         p                                   ?         @  s         t                                   C         D  w         x                                   G         H  {          |   !         !  "         "  $         $  K%         L%  &         &  '         '  (         (  *         *  O+         P+  ,         ,  -         -  .         .  0          0  S1         T1  2         2  3         3  4         4  #6         $6  W7         X7  8         8  9         9  :         :  '<         (<  [=         \=  >         >  ?         ?  @         @  +B         ,B  _C         `C  D         D  E         E  F         F  /H         ^H  I         I  J         J  K         K  ,M         .M  `N         bN  O         O  P         P  Q         Q  0S         2S  dT         fT  U         U  V         V   X         X  4Y         6Y  hZ         jZ  [         [  \         \  ^         ^  8_         :_  l`         n`  a         a  b         b  d         
d  <e         >e  pf         rf  g         g  h         h  j         j  @k         Bk  tl          2            V  ?          z  c                                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               T!         Z!  2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 i
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            <     -,-p     ,--     --    -4-    4-C-    C-v-t    v---     --C5-X    C5--    -Eb-p	    Eb--x
    -#-
    #-s-    s-ճ-\    ճ--<                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ?!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        3  8          f  k                                                          -  	          D	  +
          [
  B         r                                              4                                  &         '  M         N  t                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        V"  |#         }#  $         H  I         I  J         J  K         K  M         M  FN         GN  nO         oO  P         P  Q         Q  R         R  T         T  6U         7U  ^V         _V  W         W  X         X  Y         Y  Z         Z  &\         '\  N]         O]  v^         w^  _         _  `         `  a         a  c         c  >d         ?d  fe         ge  f         f  g         g  h         h  j         j  .k         /k  Vl         Wl  ~m         m  n         n  o         o  p         p  r         r  Fs         Gs  nt         ot  u         u  v         v  w         w  y         y  6z         7z  ^{         _{  |         |  }         }  ~         ~             &         '  N         O  v         w             ƅ         ǅ                        >         ?  f         g                        ލ         ߍ             -         .  T          U  J          }  r                      ͔                         D         E  l         m                                              4         5  \         ]                        ԣ         գ             $         %  L         M  t         u             Ī         Ū                        <                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     h            !          P!  1          1  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    n
         o
             L                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    q
         r
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            0     -,-X     ,--     -#-    #-5-x    5-}D-|    }D-w-\    w--    --h    -9b-    9b-kx-	    kx--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ?!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         #            F  3          i  V            y                                    !	          T	  X
          
                                   6         7  ]         ^                                              3         4  Z                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                        0  ,          `  \                                 $  %         %  '         '  8(         :(  j)         l)  *         *  +         +   -         -  2.         4.  d/         f/  0         0  1         1  2         2  ,4         .4  ^5         `5  6         6  7         7  8         8  &:         (:  X;         Z;  <         <  =         =  >         >   @         "@  RA         TA  B         B  C         C  D         D  F         F  LG         NG  ~H         H  I         I  K         K  BL         DL  tM         vM  N         N  O         O  
Q         Q  <R         >R  nS         pS  T         T  U         U  W         W  7X         8X  iY         jY  Z         Z  [         [  \          ]  1^         2^  c_         d_  `         `  a         a  b         b  +d         ,d  ]e         ^e  f         f  g         g  h         h  %j         &j  Wk         Xk  l         l  m         m  n         n  p          p  Qq         Rq  r         r  s         s  t         t  v         v  Kw         Lw  }x         ~x  y         y  z         z  |         |  E}         F}  w~         x~             ۀ         ܀             ?         @  q         r             Ն         ֆ             9         :  k         l             ό         Ќ             3         d                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     `            !          L!  1          1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 i
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-k;-x    k;-J-|    J-S}-\    S}-.-    .--h    -Qb-    Qb-n-	    n--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               >!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             &         '  M         N  t         u                                                           +            N  ;          q  w                                  
            =  C           E         F  l         m               !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            (         *  R         T  |         ~                                   $         &  N	         P	  x
         z
                                             "  J         L  t         v                                              F         H  p         r                                              B         D  l         n               !           -  .          C.  3/          f/  V0          0  y1          1  2         @G  iH         jH  I         I  J         J  K         K  "M         $M  LN         NN  vO         xO  P         P  Q         Q  R         R  T          T  HU         JU  rV         tV  W         W  X         X  Y         Y  [         [  D\         F\  n]         p]  ^         ^  _         _  `         `  b         b  @c         Bc  jd         ld  e         e  f         f  g         g  i         i  <j         >j  fk         hk  l         l  m           ͏         Ώ             !         "  K         L  u         v             ɖ         ʖ                        G         H  q         r             ŝ         Ɲ                        C         D  m         n                      ¤                        ?         @  i         j                                              ;                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     g            !          V!  1          2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     O
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     --    -5-x    5-}D-|    }D-w-\    w--    --h    -9b-    9b-ex-	    ex--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               =!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         #            F  4          i  W            z                                     	          T	  W
          
                                   6         7  ]         ^                                              3         4  Z                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                  P  L	          	  |
          
           $  %         %  '         '  9(         :(  k)         l)  *         *  +         +  -         -  3.         4.  e/         f/  0         0  1         1  2         2  -4         .4  _5         `5  6         6  7         7  8         8  ':         (:  Y;         Z;  <         <  =         =  >         >  !@         "@  SA         TA  B         B  C         C  D         D  F         F  MG         NG  H         H  I         I  K         K  CL         DL  uM         vM  N         N  O         O  Q         Q  =R         >R  oS         pS  T         T  U         U  W         W  6X         8X  hY         jY  Z         Z  [         [  \          ]  0^         2^  b_         d_  `         `  a         a  b         b  *d         ,d  \e         ^e  f         f  g         g  h         h  $j         &j  Vk         Xk  l         l  m         m  n         n  p          p  Pq         Rq  r         r  s         s  t         t  v         v  Jw         Lw  |x         ~x  y         y  z         z  |         |  D}         F}  v~         x~             ڀ         ܀             >         @  p         r             Ԇ         ֆ             8         :  j         l             Ό         Ќ              2         d                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     a            !          L!  1          1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 i
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-k;-x    k;-J-|    J-S}-\    S}-.-    .--h    -Wb-    Wb-n-	    n--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               >!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             &         '  M         N  t         u                                                           +            N  <          q  v                                  
            =  B           E         F  l         m               !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            )         *  S         T  }         ~                                   %         &  O	         P	  y
         z
                                   !         "  K         L  u         v                                              G         H  q         r                                              C         D  m         n               !          2  3          3  4          5  6          86  (7          [7  K8         @G  hH         jH  I         I  J         J  K         K  #M         $M  MN         NN  wO         xO  P         P  Q         Q  R         R  T          T  IU         JU  sV         tV  W         W  X         X  Y         Y  [         [  E\         F\  o]         p]  ^         ^  _         _  `         `  b         b  Ac         Bc  kd         ld  e         e  f         f  g         g  i         i  =j         >j  gk         hk  l         l  m           ̏         Ώ                       "  J         L  t         v             Ȗ         ʖ                        F         H  p         r             ĝ         Ɲ                        B         D  l         n                      ¤                        >         @  h         j                                              :                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     h            !          Z!  1          2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     N
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-m-     m--    -:-x    :-iI-|    iI-|-\    |-1-    1--h    -?b-    ?b-	-	    	--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               3!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P           %         &  L         M  s	         t	  
         
           D  j         k                                                                     5            L  (          c  ?          z  v                                     !          F!  B"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            .         0  ^         `                                               N         P  ~	         	  
         
                        >         @  n         p                                    .         0  ^         `                                               N         P  ~                                              >!         @!  n"         p"  #         #  $         $  %          &  .'         0'  ^(         `(  )         )  *         *  +         +  -          -  N.         P.  ~/         /  0         0  1         1  3         3  >4         @4  n5         p5  6         6  7         7  8          9  .:         0:  ^;         `;  <         <  =         =  >         >  @          @  NA         PA  ~B         B  C         C  D         D  F         F  >G         @G  nH         pH  I         I  J          VX  ZY          Y  Z          Z  [          
\  ]          F]  J^         d                      Ĕ             "         $  R         T                                              B         D  r         t             ҡ         ԡ             2         4  b         d             §         ħ             "         $  R         T                                              B         D  r                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     ]            !          P!  1          1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   :      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     --    -5-x    5-C-|    C-v-\    v-	-    	--h    -ub-    ub-!n-	    !n--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               <!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        3  9          f  l                                                          -  	          D	  *
          [
  A         r                                              4                                 +         ,  R         S  y                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            *         ,  V         X                                              2         4  ^	         `	  
         
                                   :         <  f         h                                              B         D  n         p                                               J         L  v         x                          !         !  &#          :  ;          *<  *=          X=  X>          >  ?          ?  @         HI  rJ         tJ  K         K  L         L  M         M  "O         $O  NP         PP  zQ         |Q  R         R  S         S  T          U  *V         ,V  VW         XW  X         X  Y         Y  Z         Z  \         \  2]         4]  ^^         `^  _         _  `         `  a         a  c         c  :d         <d  fe         he  f         f  g         g  h         h  j         j  Bk         Dk  nl         pl  m         m  n         n  o         o  q          q  Jr         Lr  vs         xs  t         t  u         u  v         v  &x         (x  Ry         Ty  ~z         z  {         {  |         |  ~         ~  .         0  Z         \                        ރ           
           6         8  b         d                                              >         @  j         l                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               <!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     O
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-y-     y--    -:-x    :-cI-|    cI-|-\    |-1-    1--h    -9b-    9b-	-	    	--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               7!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P           %         &  L         M  s	         t	  
         
           ?  e         f                                                                     0            G  $          ^  ;          u  p                                     	!          A!  <"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            /         0  _         `                                               O         P  	         	  
         
                        ?         @  o         p                                    /         0  _         `                                               O         P                                                ?!         @!  o"         p"  #         #  $         $  %          &  /'         0'  _(         `(  )         )  *         *  +         +  -          -  O.         P.  /         /  0         0  1         1  3         3  ?4         @4  o5         p5  6         6  7         7  8          9  /:         0:  _;         `;  <         <  =         =  >         >  @          @  OA         PA  B         B  C         C  D         D  F         F  ?G         @G  oH         pH  I         I  J          ^  _          _  `          `  a          6b  :c          rc  vd         d             Ô         Ĕ             #         $  S         T                                              C         D  s         t             ӡ         ԡ             3         4  c         d             ç         ħ             #         $  S         T                                              C         D  s                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     \            !          L!  1          1  ~=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   9      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     --    -5-x    5-C-|    C-v-\    v-	-    	--h    -ub-    ub-!n-	    !n--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               <!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        3  8          f  k                                                          -  	          D	  +
          [
  B         r                                              4                                  &         '  M         N  t                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            +         ,  W         X                                              3         4  _	         `	  
         
                                   ;         <  g         h                                              C         D  o         p                                               K         L  w         x                          !         !  '#          @  A          B  C          >C  >D          lD  lE          E  F         HI  sJ         tJ  K         K  L         L  M         M  #O         $O  OP         PP  {Q         |Q  R         R  S         S  T          U  +V         ,V  WW         XW  X         X  Y         Y  Z         Z  \         \  3]         4]  _^         `^  _         _  `         `  a         a  c         c  ;d         <d  ge         he  f         f  g         g  h         h  j         j  Ck         Dk  ol         pl  m         m  n         n  o         o  q          q  Kr         Lr  ws         xs  t         t  u         u  v         v  'x         (x  Sy         Ty  z         z  {         {  |         |  ~         ~  /         0  [         \                        ߃                      7         8  c         d                                              ?         @  k         l                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               <!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     N
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-q;-x    q;-J-|    J-Y}-\    Y}--    --h    -]b-    ]b--	    --h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               C!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             &         '  M         N  t         u                       r  n                                              >  :          q  ^                                                           E         F  l         m               !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ,         .  Z         \                                              @         B  n	         p	  
         
                        &         (  T         V                                              :         <  h         j                                             "  N         P  |         ~                        !         !  4"         6"  b#         d#  $         $  %         %  &         &  (         (  H)         J)  v*         x*  +         +  ,         ,   .         .  ./         0/  \0         ^0  1         1  2         2  3         3  5         5  B6         D6  p7         r7  8         8  9         9  :         :  (<         *<  V=         X=  >         >  ?         ?  @         @  B         B  <C         >C  jD         lD  E         E  F          `  a          a  b          c  d          <d  0e          he  \f         l  m         m  o          o  Mp         Np  {q         |q  r         r  s         s  u         u  3v         4v  aw         bw  x         x  y         y  z         z  |         |  G}         H}  u~         v~             р         Ҁ              -         .  [         \                                              A         B  o         p             ˍ         ̍             '                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               /!         >!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   :      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-e;-x    e;-I-|    I-M}-\    M}-m-    m--h    -{b-    {b-x-	    x--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               @!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 #	         $	  J
         K
  q                                                      !  G          H  B            y                                  $            [  H          ~  k                          !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            -         .  [         \                                              A         B  o	         p	  
         
                        '         (  U         V                                              ;         <  i         j                                   !         "  O         P  }         ~                        !         !  5"         6"  c#         d#  $         $  %         %  &         &  (         (  H)         J)  v*         x*  +         +  ,         ,   .         .  ./         0/  \0         ^0  1         1  2         2  3         3  5         5  B6         D6  p7         r7  8         8  9         9  :         :  (<         *<  V=         X=  >         >  ?         ?  @         @  B         B  <C         >C  jD         lD  E         E  F         F  G         G  "I         $I  PJ         RJ  ~K         K  L         L  M         M  O         
O  6P         8P  dQ         fQ  R         R  S         S  T         T  V         V  JW         LW  xX         zX  Y         Y  Z         Z  \         \  0]         2]  ^^         `^  _         _  `         `  a         a  c         c  Dd         Fd  re         te  f         f  g         g  h          pl  ^m          m  n          n  o          o  p           q  r         ʐ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               =!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 }
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-q;-x    q;-J-|    J-Y}-\    Y}--    --h    -]b-    ]b--	    --h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               C!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             &         '  M         N  t         u                       r  m                                              >  9          q  _                                                           E         F  l         m               !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            -         .  [         \                                              A         B  o	         p	  
         
                        '         (  U         V                                              ;         <  i         j                                   !         "  O         P  }         ~                        !         !  5"         6"  c#         d#  $         $  %         %  &         &  (         (  I)         J)  w*         x*  +         +  ,         ,  .         .  //         0/  ]0         ^0  1         1  2         2  3         3  5         5  C6         D6  q7         r7  8         8  9         9  :         :  )<         *<  W=         X=  >         >  ?         ?  @         @  B         B  =C         >C  kD         lD  E         E  F          f  g          g  h          h  i          j  k          Dk  8l         l  m         m  o          o  Lp         Np  zq         |q  r         r  s         s  u         u  2v         4v  `w         bw  x         x  y         y  z         z  |         |  F}         H}  t~         v~             Ѐ         Ҁ              ,         .  Z         \                                              @         B  n         p             ʍ         ̍             &                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               /!         >!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   9      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-e;-x    e;-I-|    I-M}-\    M}-m-    m--h    -b-    b-x-	    x--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               <!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 #	         $	  J
         K
  q                                                      !  G          H  A            x                                  $            [  I          ~  l                          !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ,         .  Z         \                                              @         B  n	         p	  
         
                        &         (  T         V                                              :         <  h         j                                             "  N         P  |         ~                        !         !  4"         6"  b#         d#  $         $  %         %  &         &  (         (  I)         J)  w*         x*  +         +  ,         ,  .         .  //         0/  ]0         ^0  1         1  2         2  3         3  5         5  C6         D6  q7         r7  8         8  9         9  :         :  )<         *<  W=         X=  >         >  ?         ?  @         @  B         B  =C         >C  kD         lD  E         E  F         F  G         G  #I         $I  QJ         RJ  K         K  L         L  M         M  	O         
O  7P         8P  eQ         fQ  R         R  S         S  T         T  V         V  KW         LW  yX         zX  Y         Y  Z         Z  \         \  1]         2]  _^         `^  _         _  `         `  a         a  c         c  Ed         Fd  se         te  f         f  g         g  h          Lr  :s          xs  ft          t  u          u  v          v  w         ʐ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ?!         R!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 }
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -5-    5-:-    :-uI-    uI-|-d    |-k
-    k
-	-0    	--    -b-H	    b-c-P
    c-U-
    U-9׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               D!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P           %         &  L         M  s	         t	  
         
           D  j         k                                                                     5            L  (          c  ?          z  v                                     !          F!  B"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q   w  operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       \            *         ,  V         X                                              2         4  ^	         `	  
         
                                   :         <  f         h                                              B         D  n         p                                               J         L  v         x                          !         !  &#         (#  R$         T$  ~%         %  &         &  '         '  )         )  .*         0*  Z+         \+  ,         ,  -         -  .         .  
0         0  61         81  b2         d2  3         3  4         4  5         5  7         7  >8         @8  j9         l9  :         :  ;         ;  <         <  >         >  F?         H?  r@         t@  A         A  B         B  C          h  i          *j  #k          Vk  Ol          l  {m          m  n         n  p         p  0q         2q  \r         ^r  s         s  t         t  u         u  w         w  8x         :x  dy         fy  z         z  {         {  |         |  ~         ~  @         B  l         n                      Ă                        C         E  o         q             ǉ         ɉ                                b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               D!         \!  1         
2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 y
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-5-x    5-C-|    C-v-\    v-i0-    i0--h    -ub-    ub-x-	    x--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               B!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        3  9          f  l                                                          -  	          D	  *
          [
  A         r                                              4                                 +         ,  R         S  y                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                     &  '          X  Y                                 %  	'         
'  9(         :(  i)         j)  *         *  +         +  ,         ,  ).         *.  Y/         Z/  0         0  1         1  2         2  4         4  I5         J5  y6         z6  7         7  8         8  	:         
:  9;         :;  i<         j<  =         =  >         >  ?         ?  )A         *A  YB         ZB  C         C  D         D  E         E  G         G  IH         JH  yI         h  i         i  *k         ,k  Zl         \l  m         m  n         n  o         o  q         q  Jr         Lr  zs         |s  t         t  u         u  
w         w  :x         <x  jy         ly  z         z  {         {  |         |  *~         ,~  Z         \                                              J         L  z         |             ڈ         ܈  
           :         <  j         P  ~                      ސ                      >         @  n         p             Ζ         Ж              .         0  ^         `                                               N         P  ~                      ޣ                      >         @  n         p             Ω         Щ              .         0  ^         `                                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               :!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 }
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -5-    5-:-    :-oI-    oI-|-d    |-q
-    q
--0    --    -b-H	    b-c-P
    c-U-
    U-9׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               C!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P           %         &  L         M  s	         t	  
         
           ?  e         f                                                                     0            G  $          ^  ;          u  p                                     	!          A!  <"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q   w  operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       \            +         ,  W         X                                              3         4  _	         `	  
         
                                   ;         <  g         h                                              C         D  o         p                                               K         L  w         x                          !         !  '#         (#  S$         T$  %         %  &         &  '         '  )         )  /*         0*  [+         \+  ,         ,  -         -  .         .  0         0  71         81  c2         d2  3         3  4         4  5         5  7         7  ?8         @8  k9         l9  :         :  ;         ;  <         <  >         >  G?         H?  s@         t@  A         A  B         B  C         h  )j         *j  Uk         Vk  l         l  m         m  n          n  o          p  p          2q  +r          ^r  Ws          s  t         t  u         u  w         w  9x         :x  ey         fy  z         z  {         {  |         |  ~         ~  A         B  m         n             Â         Ă                        D         E  p         q             ȉ         ɉ                                b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               B!         X!  1         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 y
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-5-x    5-C-|    C-v-\    v-i0-    i0--h    -ub-    ub-x-	    x--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               >!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        3  8          f  k                                                          -  	          D	  +
          [
  B         r                                              4                                  &         '  M         N  t                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                        !          R  S                                 %  '         
'  8(         :(  h)         j)  *         *  +         +  ,         ,  (.         *.  X/         Z/  0         0  1         1  2         2  4         4  H5         J5  x6         z6  7         7  8         8  :         
:  8;         :;  h<         j<  =         =  >         >  ?         ?  (A         *A  XB         ZB  C         C  D         D  E         E  G         G  HH         JH  xI         h  i         i  +k         ,k  [l         \l  m         m  n         n  o         o  q         q  Kr         Lr  {s         |s  t         t  u         u  w         w  ;x         <x  ky         ly  z         z  {         {  |         |  +~         ,~  [         \                                              K         L  {         |             ۈ         ܈             ;         <  k         P                        ߐ                      ?         @  o         p             ϖ         Ж              /         0  _         `                                               O         P                        ߣ                      ?         @  o         p             ϩ         Щ              /         0  _         `                                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               :!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 }
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -/-    /-};-    };-J-    J-e}-d    e}-G-    G--p    -b-    b-'-
    '-5-
    5--t    -9׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               A!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             &         '  M         N  t         u                       r  n                                              >  :          q  ^                                                           E         F  l         m               !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         X  a                                              H  Q         %  1&         2&  ['         \'  (         (  )         )  *         *  ,         ,  --         .-  W.         X.  /         /  0         0  1         1  2          3  )4         *4  S5         T5  }6         ~6  7         7  8         8  9         9  %;         &;  O<         P<  y=         z=  >         >  ?         ?  @         @  !B         "B  KC         LC  uD         vD  E         E  F         F  G         G  I         I  FJ         HJ  pK         rK  L         L  M         M  N         N  P         P  BQ         DQ  lR         nR  S         S  T         T  U         U  W         W  >X         @X  hY         jY  Z         Z  [         [  \         \  ^         ^  :_         <_  d`         f`  a         a  b         b  c         c  e         e  6f         8f  `g         bg  h         h  i         i  j         j  l         
l  2m         4m  \n         ^n  o         o  p         p  q         q  s         s  .t         0t  Xu         Zu  v         v  w         w  x         x   z         z  *{         ,{  T|         V|  ~}         }  ~         ~                        &         (  P         R  z         |             Ά         І             "         $  L         N  v         x             ʍ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               M!         f!   2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    n
         o
             L      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     H
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-y-     y--    -q;-x    q;-J-|    J-Y}-\    Y}-C-    C--h    -{b-    {b-낳-	    낳--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               6!         @!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 #	         $	  J
         K
  q                                                      !  G          H  B            y                                  $            [  H          ~  k                          !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            1         2  c         d                                   +         ,  ]         ^  	         	  
         
             %         &  W         X                                               Q         R                                              K         L  }         ~                                   E          F   w!         x!  "         "  #          P<  D=          z=  n>          >  ?          ?  @          @  A         G  H         H  I         I  -K         .K  _L         `L  M         M  N         N  O         O  'Q         (Q  YR         ZR  S         S  T         T  U         U  !W         "W  SX         TX  Y         Y  Z         Z  [         [  ]         ]  M^         N^  _         _  `         `  a         a  c         c  Gd         Hd  ye         ze  f         f  g         g  i         i  Aj         Bj  sk         tk  l         l  m         Pp  q         q  r         r  s         s  u         u  Hv         Jv  zw         |w  x         x  y         y  {         {  B|         D|  t}         v}  ~         ~             
           <         >  n         p             ҅         ԅ             6         8  h         j             ̋         ΋              0         2  b         d             Ƒ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               9!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -/-    /-};-    };-J-    J-e}-d    e}-G-    G--p    -b-    b-'-
    '-5-
    5--t    -9׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               A!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             &         '  M         N  t         u                       r  m                                              >  9          q  _                                                           E         F  l         m               !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                             8  A          t  }         %  0&         2&  Z'         \'  (         (  )         )  *         *  ,         ,  ,-         .-  V.         X.  /         /  0         0  1         1  2          3  (4         *4  R5         T5  |6         ~6  7         7  8         8  9         9  $;         &;  N<         P<  x=         z=  >         >  ?         ?  @         @   B         "B  JC         LC  tD         vD  E         E  F         F  G         G  I         I  GJ         HJ  qK         rK  L         L  M         M  N         N  P         P  CQ         DQ  mR         nR  S         S  T         T  U         U  W         W  ?X         @X  iY         jY  Z         Z  [         [  \         \  ^         ^  ;_         <_  e`         f`  a         a  b         b  c         c  e         e  7f         8f  ag         bg  h         h  i         i  j         j  	l         
l  3m         4m  ]n         ^n  o         o  p         p  q         q  s         s  /t         0t  Yu         Zu  v         v  w         w  x         x  z         z  +{         ,{  U|         V|  }         }  ~         ~                        '         (  Q         R  {         |             φ         І             #         $  M         N  w         x             ˍ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               M!         f!   2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    n
         o
             L      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     I
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     --    -q;-x    q;-J-|    J-Y}-\    Y}-C-    C--h    -ub-    ub-낳-	    낳--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               8!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 #	         $	  J
         K
  q                                                      !  G          H  A            x                                  $            [  I          ~  l                          !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            0         2  b         d                                   *         ,  \         ^  	         	  
         
             $         &  V         X                                               P         R                                              J         L  |         ~                                   D          F   v!         x!  "         "  #          "B  C          LC  @D          vD  jE          E  F         G  H         H  I         I  ,K         .K  ^L         `L  M         M  N         N  O         O  &Q         (Q  XR         ZR  S         S  T         T  U         U   W         "W  RX         TX  Y         Y  Z         Z  [         [  ]         ]  L^         N^  ~_         _  `         `  a         a  c         c  Fd         Hd  xe         ze  f         f  g         g  i         i  @j         Bj  rk         tk  l         l  m          ^n  Ro         Pp  q         q  r         r  s         s  u         u  Iv         Jv  {w         |w  x         x  y         y  {         {  C|         D|  u}         v}  ~         ~                        =         >  o         p             Ӆ         ԅ             7         8  i         j             ͋         ΋              1         2  c         d             Ǒ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               5!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     -;-    ;-6-    6-D-    D-x-l    x-72-    72--x    -b-	    b---
    --;-
    ;--|    -ճ-    ճ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               C!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         #            F  3          i  V            y                                    	          T	  U
          
                                   6         7  ]         ^                                   !         "  O         P  }                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |        $  %          0?  =@          s@  A          A  B          B  D          <D  IE         K  L         L  N         N  EO         FO  mP         nP  Q         Q  R         R  S         S  U         U  5V         6V  ]W         ^W  X         X  Y         Y  Z         Z  [         [  %]         &]  M^         N^  u_         v_  `         `  a         a  b         b  d         d  =e         >e  ef         ff  g         g  h         h  i         i  k         k  -l         .l  Um         Vm  }n         ~n  o         o  p         p  q         q  s         s  Et         Ft  mu         nu  v         v  w         w  x         x  z         z  5{         6{  ]|         ^|  }         }  ~         ~                        %         &  M         N  u         v             ņ         Ɔ                        =         >  e         f                        ݎ         ގ             -         .  T         V  |         ~             ̕         Ε                        D         F  l         n                                              4         6  \         ^                        Ԥ         ֤             $         &  L         N  t         v             ī         ƫ                        <         >  d         f                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               D!         T!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    n
         o
             L      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    n
         o
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     T
            0     -,-X     ,--     -M-    M-;-x    ;-;J-|    ;J-}-\    }--    --h    -ib-    ib-낳-	    낳--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               N!         V!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             &         '  M         N  t         u                                                           +            N  ;          q  r                                  
            =  >         B  h         i                          !         !  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            2         4  f         h                                   6         8  j         l  	         	  
         
             :         <  n         p                        
           >         @  r         t                                   B         D  v         x                                   F         H  z          |   !         !  "         "  $         $  %          &  3'         4'  g(         h(  )         )  *         *  ,         ,  7-         8-  k.         l.  /         /  0         0  2         2  ;3         <3  o4         p4  5         5  6         6  8         8  ?9         @9  s:         t:  ;         ;  <         <  >         >  C?         D?  w@         x@  A         A  B         B  D         D  GE         HE  {F         |F  G         G  H          K  L          M  N          TN  aO          O  P          P  Q         "o  Tp         Vp  q         q  r         r  s         s  $u         &u  Xv         Zv  w         w  x         x  y         y  ({         *{  \|         ^|  }         }  ~         ~             ,         .  `         b             Ȅ         ʄ             0         2  d         f             ̊         Ί              4         6  h         j             А         Ґ             8                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               2!         :!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     -;-    ;-6-    6-D-    D-x-l    x-72-    72--x    -b-	    b---
    --;-
    ;--|    -ճ-    ճ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               C!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         #            F  4          i  W            z                                    	          T	  V
          
                                   6         7  ]         ^                                              K         L  x                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |        $  %          E  F          F  G          H  I          HI  UJ          J  K         K  L         L  N         N  DO         FO  lP         nP  Q         Q  R         R  S         S  U         U  4V         6V  \W         ^W  X         X  Y         Y  Z         Z  [         [  $]         &]  L^         N^  t_         v_  `         `  a         a  b         b  d         d  <e         >e  df         ff  g         g  h         h  i         i  k         k  ,l         .l  Tm         Vm  |n         ~n  o         o  p         p  q         q  s         s  Dt         Ft  lu         nu  v         v  w         w  x         x  z         z  4{         6{  \|         ^|  }         }  ~         ~                        $         &  L         N  t         v             Ć         Ɔ                        <         >  d         f                        ܎         ގ             ,         .  U         V  }         ~             ͕         Ε                        E         F  m         n                                              5         6  ]         ^                        դ         ֤             %         &  M         N  u         v             ū         ƫ                        =         >  e         f                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               B!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    n
         o
             L      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    n
         o
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     S
            0     -,-X     ,--     -G-    G-;-x    ;-5J-|    5J-}-\    }--    --h    -cb-    cb-낳-	    낳--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               L!         V!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             &         '  M         N  t         u                                                           +            N  <          q  q                                  
  
          =  =         =  c         d                          !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            3         4  g         h                                   7         8  k         l  	         	  
         
             ;         <  o         p                                   ?         @  s         t                                   C         D  w         x                                   G         H  {          |   !         !  "         "  $         $  %          &  2'         4'  f(         h(  )         )  *         *  ,         ,  6-         8-  j.         l.  /         /  0         0  2         2  :3         <3  n4         p4  5         5  6         6  
8         8  >9         @9  r:         t:  ;         ;  <         <  >         >  B?         D?  v@         x@  A         A  B         B  D         D  FE         HE  zF         |F  G         G  H          R  *S          `S  mT          T  U          U  V          )W  6X         "o  Up         Vp  q         q  r         r  s         s  %u         &u  Yv         Zv  w         w  x         x  y         y  ){         *{  ]|         ^|  }         }  ~         ~             -         .  a         b             Ʉ         ʄ             1         2  e         f             ͊         Ί             5         6  i         j             ѐ         Ґ             9                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               0!         :!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     ---X     ---     -e-    e-:-x    :-I-|    I-|-\    |--    --h    -b-    b-wx-	    wx--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               [!         \!  	2         
2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P           %         &  L         M  s	         t	  
         
           D  j         k                                                         %            C            a  9            W                                              6   '!          i!  Z"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            7         8  o         p                                   O         P             	         	  
         
  /         0  g         h                                   G         H                                   '         (  _         `                                   ?         @  w         x                          "          "  W#         X#  $         $  %         %  &          '  6(         8(  n)         p)  *         *  +         +  -         -  N.         P.  /         /  0         0  1         1  .3         03  f4         h4  5         5  6         6  8         8  F9         H9  ~:         :  ;         ;  <         <  &>         (>  ^?         `?  @         @  A         A  C         C  >D         @D  vE         xE  F         F  G         G  I         I  K         K  ML         NL  M         M  N         N  O         O  -Q         .Q  eR         fR  S         S  T         T  V         V  EW         FW  }X         ~X  Y         Y  Z         Z  %\         &\  ]]         ^]  ^         ^  _         _  a         a  =b         >b  uc         vc  d         d  e         e  g         g  Uh         Vh  i         i  j         j  k         k  5m         6m  mn          ~n  `o          o  p          p  q          q  r          s   t                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               T!         Z!  2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 i
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     ---`     ---     -_-    _-};-    };-J-    J-e}-d    e}-q-    q--0    --    -b-H	    b-c-P
    c-O-
    O-E׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               V!         X!  2         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 #	         $	  J
         K
  q                                                      !  G          H  9            p                                  $            [  H          ~  k                          !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q   w  operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       \            5         6  k         l                                   C         D  y         z  	         	  
         
             Q         R                                   )         *  _         `                                   7         8  m         n                                   E         F  {         |               !         !  #         #  S$         T$  %         %  &         &  '         '  +)         ,)  a*         b*  +         +  ,         ,  .         .  9/         :/  o0         p0  1         1  2         2  4         4  G5         H5  }6         ~6  7         7  8         8  :          :  U;         V;  <         <  =         =  >         >  -@         .@  cA         dA  B         B  C         C  E         E  ;F         <F  pG         rG  H         H  I         nn  o         o  p         p  r         r  Ds         Fs  zt         |t  u         u  v          v  w          x  !y          Ty  Wz          z  {          {  |         |  *~         ,~  `         b             ́         ΁             8         :  n         p             ڇ         ܇             F         H  |         ~                                           b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               J!         X!  1         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 m
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     ---X     ---     -_-    _-:-x    :-I-|    I-|-\    |-U-    U--h    -b-    b-wx-	    wx--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               W!         X!  2         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P           %         &  L         M  s	         t	  
         
           ?  e         f                                                                     <            Y  1          v  N                                              ,   !          _!  O"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            6         8  n         p                                   N         P             	         	  
         
  .         0  f         h                                   F         H  ~                                 &         (  ^         `                                   >         @  v         x                          "          "  V#         X#  $         $  %         %  &          '  7(         8(  o)         p)  *         *  +         +  -         -  O.         P.  /         /  0         0  1         1  /3         03  g4         h4  5         5  6         6  8         8  G9         H9  :         :  ;         ;  <         <  '>         (>  _?         `?  @         @  A         A  C         C  ?D         @D  wE         xE  F         F  G         G  I         I  K         K  LL         NL  M         M  N         N  O         O  ,Q         .Q  dR         fR  S         S  T         T  V         V  DW         FW  |X         ~X  Y         Y  Z         Z  $\         &\  \]         ^]  ^         ^  _         _  a         a  <b         >b  tc         vc  d         d  e         e  g         g  Th         Vh  i         i  j         j  k         k  4m         6m  ln          Ft  (u          nu  Pv          v  xw          w  x          x  y                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               R!         Z!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 i
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     ---`     ---     -Y-    Y-};-    };-J-    J-e}-d    e}-q-    q--0    --    -b-H	    b-c-P
    c-I-
    I-E׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               U!         V!   2         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 #	         $	  J
         K
  q                                                      !  G          H  :            q                                  $            [  I          ~  l                          !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q   w  operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       \            4         6  j         l                                   B         D  x         z  	         	  
         
             P         R                                   (         *  ^         `                                    6         8  l         n                                   D         F  z         |               !         !  #         #  R$         T$  %         %  &         &  '         '  *)         ,)  `*         b*  +         +  ,         ,  .         .  8/         :/  n0         p0  1         1  2         2  4         4  F5         H5  |6         ~6  7         7  8         8  :          :  T;         V;  <         <  =         =  >         >  ,@         .@  bA         dA  B         B  C         C  E         E  :F         <F  qG         rG  H         H  I         nn  o         o  p         p  r         r  Es         Fs  {t         |t  u         u  v         v  x         x  Sy         Ty  z         z  {         {  |          |  }          ,~  /          b  e                      ΁  т           9         :  o         p             ۇ         ܇             G         H  }         ~                                           b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               H!         T!  1         1  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 m
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     ---X     ---     -_-    _-5-x    5-D-|    D-w-\    w-7-    7--h    -b-    b-x-	    x--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               [!         \!  	2         
2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         #            F  3          i  V            y                                    	          T	  P
          
                                   6         7  ]         ^                                              3         4  Z                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         T  :          v  \            ~                                 U$  %         %  &         &  '         '  ,)         -)  b*         c*  +         +  ,         ,  .         .  :/         ;/  p0         q0  1         1  2         2  4         4  H5         I5  ~6         6  7         7  8         8   :         !:  V;         W;  <         <  =         =  >         >  .@         /@  dA         eA  B         B  C         C  E         E  <F         =F  rG         G  I         I  CJ         EJ  yK         {K  L         L  M         M  O         O  QP         SP  Q         Q  R         R  S         S  )U         +U  _V         aV  W         W  X         X  Z         Z  7[         9[  m\         o\  ]         ]  ^         ^  `         `  Ea         Ga  {b         }b  c         c  d         d  f         f  Sg         Ug  h         h  i         i  j         j  +l         l  
n         ړ             E         F  {         |                                   S         T                                   +         ,  a         b             ͤ         Τ             9         :  o         p             ۪         ܪ             G         H  }         ~                                    U         V                      µ             -                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               K!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 i
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     ---`     ---     -e-    e-;-    ;-J-    J-k}-d    k}--    -%-p    %-b-    b-3-
    3-A-
    A-հ-t    հ-E׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               X!         `!  2         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             &         '  M         N  t         u                                                           +            N  ;          q  o                                  
            =  ;           E         F  l         m               !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                 D  7          j  ]                     T#  $         $  %         %  &         &  3(         4(  k)         l)  *         *  +         +  -         -  K.         L.  /         /  0         0  1         1  +3         ,3  c4         d4  5         5  6         6  8         8  C9         D9  {:         |:  ;         ;  <         <  #>         $>  [?         \?  @         @  A         A  C         C  ;D         <D  sE         tE  F         F  G         G  I         I  RJ         TJ  K         K  L         L  M         M  2O         4O  jP         lP  Q         Q  R         R  T         T  JU         LU  V         V  W         W  X         X  *Z         ,Z  b[         d[  \         \  ]         ]  
_         _  B`         D`  za         |a  b         b  c         c  "e         $e  Zf         \f  g         g  h         h  j         j  :k         <k  rl         tl  m         m  n         n  p         p  Rq         Tq  r         r  s         s  t         t  2v         4v  jw         lw  x         x  y         y  {         {  J|         L|  }         }  ~         ~             *         ,  b         d             ҄         Ԅ  
           B         D  z         |                        "         $  Z         \             ʏ         ̏                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               Y!         j!  2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    n
         o
             L      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     &
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     ---X     ---     -Y-    Y-5-x    5-D-|    D-w-\    w-7-    7--h    -b-    b-x-	    x--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               W!         X!  2         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         #            F  4          i  W            z                                    	          T	  O
          
                                   6         7  ]         ^                                              3         4  Z                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                  B  (          d  J            l         U$  %         %  &         &  '         '  +)         -)  a*         c*  +         +  ,         ,  .         .  9/         ;/  o0         q0  1         1  2         2  4         4  G5         I5  }6         6  7         7  8         8  :         !:  U;         W;  <         <  =         =  >         >  -@         /@  cA         eA  B         B  C         C  E         E  ;F         =F  qG         G  I         I  DJ         EJ  zK         {K  L         L  M         M  O         O  RP         SP  Q         Q  R         R  S         S  *U         +U  `V         aV  W         W  X         X  Z         Z  8[         9[  n\         o\  ]         ]  ^         ^  `         `  Fa         Ga  |b         }b  c         c  d         d  f         f  Tg         Ug  h         h  i         i  j         j  ,l         l  	n         ړ             D         F  z         |                                   R         T                                   *         ,  `         b             ̤         Τ             8         :  n         p             ڪ         ܪ             F         H  |         ~                                    T         V                      µ             ,                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               K!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 i
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     ---`     ---     -_-    _-;-    ;-J-    J-k}-d    k}--    -%-p    %-b-    b-3-
    3-A-
    A-հ-t    հ-E׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               V!         \!  2         
2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             &         '  M         N  t         u                                                           +            N  <          q  n                                  
            =  :           E         F  l         m               !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                             (   !          N!  A"         T#  $         $  %         %  &         &  2(         4(  j)         l)  *         *  +         +  -         -  J.         L.  /         /  0         0  1         1  *3         ,3  b4         d4  5         5  6         6  
8         8  B9         D9  z:         |:  ;         ;  <         <  ">         $>  Z?         \?  @         @  A         A  C         C  :D         <D  rE         tE  F         F  G         G  I         I  SJ         TJ  K         K  L         L  M         M  3O         4O  kP         lP  Q         Q  R         R  T         T  KU         LU  V         V  W         W  X         X  +Z         ,Z  c[         d[  \         \  ]         ]  _         _  C`         D`  {a         |a  b         b  c         c  #e         $e  [f         \f  g         g  h         h  j         j  ;k         <k  sl         tl  m         m  n         n  p         p  Sq         Tq  r         r  s         s  t         t  3v         4v  kw         lw  x         x  y         y  {         {  K|         L|  }         }  ~         ~             +         ,  c         d             ӄ         Ԅ             C         D  {         |                        #         $  [         \             ˏ         ̏                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               Y!         j!  2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    n
         o
             L      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     %
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -G-    G-:-x    :-I-|    I-|-\    |-;-    ;--h    -ub-    ub-!-	    !--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               M!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P           %         &  L         M  s	         t	  
         
           D  j         k                                                                     5            L  0          c  G          z  ~                                     !          F!  J"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            2         4  f         h                                   6         8  j         l  	         	  
         
             :         <  n         p                        
           >         @  r         t                                   B         D  v         x                                   F         H  z          |   !         !  "         "  $         $  J%         L%  ~&         &  '         '  (         (  *         *  N+         P+  ,         ,  -         -  .         .  0          0  R1         T1  2         2  3         3  4         4  "6         $6  V7         X7  8         8  9         9  :         :  &<         (<  Z=         \=  >         >  ?         ?  @         @  *B         ,B  ^C         `C  D         D  E         E  F         F  .H         0H  bI         dI  J         J  K         K  L          M  2N         4N  fO         hO  P         P  Q         Q  S         S  6T         8T  jU         lU  V         W  QX         SX  Y         Y  Z         Z  [         [  !]         #]  U^         W^  _         _  `         `  a         a  %c         'c  Yd         [d  e         e  f         f  g         g  )i         +i  ]j            ƅ            	          <  L                        Ҋ         T                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               4!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   B      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -M-    M-5-x    5-C-|    C-v-\    v-I-    I--h    -b-    b-3n-	    3n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               P!         X!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         3  1          f  d                                                          -  	          D	  "
          [
  9         r                                              4                                 +         ,  R         S  y                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            &         (  N         P  v         x                                              >	         @	  f
         h
                                              .         0  V         X  ~                                                        F         H  n         p                                              6         8  ^          `   !         !  "         "  $         $  :%         <%  b&         d&  '         '  (         (  )         )  +         +  *,         ,,  R-         T-  z.         |.  /         /  0         0  1         1  3         3  B4         D4  j5         l5  6         6  7         7  8         8  
:         :  2;         4;  Z<         \<  =         =  >         >  ?         ?  @         @  "B         $B  JC         LC  rD         tD  E         E  F          F  G          G  H          (I  $J          ZJ  VK          K  L         Bm  hn         jn  o         o  p         p  q         q  s         
s  0t         2t  Xu         Zu  v         v  w         w  x         x  y         y   {         "{  H|         J|  p}         r}  ~         ~                                   8         :  `         b                        ؇         ڇ              (         *  P         R  x         z             Ȏ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               N!         X!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     O
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -G-    G-:-x    :-{I-|    {I-|-\    |-;-    ;--h    -ub-    ub-!-	    !--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               M!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P           %         &  L         M  s	         t	  
         
           ?  e         f                                                                     0            G  ,          ^  C          u  x                                     !          A!  D"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            3         4  g         h                                   7         8  k         l  	         	  
         
             ;         <  o         p                                   ?         @  s         t                                   C         D  w         x                                   G         H  {          |   !         !  "         "  $         $  K%         L%  &         &  '         '  (         (  *         *  O+         P+  ,         ,  -         -  .         .  0          0  S1         T1  2         2  3         3  4         4  #6         $6  W7         X7  8         8  9         9  :         :  '<         (<  [=         \=  >         >  ?         ?  @         @  +B         ,B  _C         `C  D         D  E         E  F         F  /H         0H  cI         dI  J         J  K         K  L          M  3N         4N  gO         hO  P         P  Q         Q  S         S  7T         8T  kU         lU  V         W  RX         SX  Y         Y  Z         Z  [         [  "]         #]  V^         W^  _         _  `         `  a         a  &c         'c  Zd         [d  e         e  f         f  g         g  *i         +i  ^j                      H  X                      Ύ  ޏ            !         T                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               4!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   A      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -G-    G-5-x    5-C-|    C-v-\    v-I-    I--h    -b-    b--n-	    -n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               N!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         3  0          f  c                                                          -  	          D	  #
          [
  :         r                                              4                                  &         '  M         N  t                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            '         (  O         P  w         x                                              ?	         @	  g
         h
                                              /         0  W         X                                                          G         H  o         p                                              7         8  _          `   !         !  "         "  $         $  ;%         <%  c&         d&  '         '  (         (  )         )  +         +  +,         ,,  S-         T-  {.         |.  /         /  0         0  1         1  3         3  C4         D4  k5         l5  6         6  7         7  8         8  :         :  3;         4;  [<         \<  =         =  >         >  ?         ?  @         @  #B         $B  KC         LC  sD         tD  E         E  F          L  M          M  N          "O  P          TP  PQ          Q  R         Bm  in         jn  o         o  p         p  q         q  	s         
s  1t         2t  Yu         Zu  v         v  w         w  x         x  y         y  !{         "{  I|         J|  q}         r}  ~         ~                                   9         :  a         b                        ه         ڇ             )         *  Q         R  y         z             Ɏ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               L!         T!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     N
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -A-    A-};-x    };-J-|    J-e}-\    e}--    --h    -Wb-    Wb-!-	    !--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               K!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             &         '  M         N  t         u                       r  v                                              >  B          q  ^                                                           E         F  l         m               !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            0         2  b         d                                   *         ,  \         ^  	         	  
         
             $         &  V         X                                               P         R                                              J         L  |         ~                                   D          F   v!         x!  "         "  #          $  %          <%  4&          d&  \'          '  (          (  )         F  G         G  &I         (I  XJ         ZJ  K         K  L         L  M         M   O         "O  RP         TP  Q         Q  R         R  S         S  U         U  LV         NV  ~W         W  X         X  Y         Y  [         [  F\         H\  x]         z]  ^         ^  _         _  a         a  @b         Bb  rc         tc  d         d  e         e  g         
g  :h         <h  li         ni  j         j  k         k  m         m  5n         6n  go         ho  p         p  q         q  r         r  /t         0t  au         bu  v         v  w         w  x         x  )z         *z  [{         \{  |         |  }         }  ~         ~  #         $  U         V                                              O         P                                              I         J  {                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               +!         2!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   B      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -G-    G-q;-x    q;-J-|    J-Y}-\    Y}-o-    o--h    -b-    b-x-	    x--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               J!         X!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 #	         $	  J
         K
  q                                                      !  G          H  J                                              $  &          [  H          ~  k                          !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            )         *  S         T  }         ~                                   %         &  O	         P	  y
         z
                                   !         "  K         L  u         v                                              G         H  q         r                                              C         D  m         n               !         !  "         $  <%         >%  f&         h&  '         '  (         (  )         )  +         +  8,         :,  b-         d-  .         .  /         /  0         0  
2         2  43         63  ^4         `4  5         5  6         6  7         7  9         9  0:         2:  Z;         \;  <         <  =         =  >         >  @         @  ,A         .A  VB         XB  C         C  D         D  E         E  F          G  (H         :H  bI         dI  J         J  K         K  L         L  
N         N  4O         6O  ^P         `P  Q         Q  R         R  S         S  U         U  0V         2V  ZW         \W  X         X  Y         Y  Z         Z  \         \  ,]         .]  V^         X^  _         _  `         `  a         a  b          c  (d         *d  Re         Te  |f         ~f  g         g  h          ~u  v          v  w          w  x          ,y  0z          fz  j{         ʎ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               K!         ^!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 }
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-};-x    };-J-|    J-e}-\    e}--    --h    -]b-    ]b-!-	    !--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               G!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             &         '  M         N  t         u                       r  u                                              >  A          q  _                                                           E         F  l         m               !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            1         2  c         d                                   +         ,  ]         ^  	         	  
         
             %         &  W         X                                               Q         R                                              K         L  }         ~                                   E          F   w!         x!  "         "  #          )  *          +  +          ,,  $-          T-  L.          |.  t/         F  G         G  'I         (I  YJ         ZJ  K         K  L         L  M         M  !O         "O  SP         TP  Q         Q  R         R  S         S  U         U  MV         NV  W         W  X         X  Y         Y  [         [  G\         H\  y]         z]  ^         ^  _         _  a         a  Ab         Bb  sc         tc  d         d  e         e  	g         
g  ;h         <h  mi         ni  j         j  k         k  m         m  4n         6n  fo         ho  p         p  q         q  r         r  .t         0t  `u         bu  v         v  w         w  x         x  (z         *z  Z{         \{  |         |  }         }  ~         ~  "         $  T         V                                              N         P                                              H         J  z                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               -!         6!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   A      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -A-    A-q;-x    q;-J-|    J-Y}-\    Y}-o-    o--h    -b-    b-x-	    x--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               H!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 #	         $	  J
         K
  q                                                      !  G          H  I                                              $  %          [  I          ~  l                          !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            (         *  R         T  |         ~                                   $         &  N	         P	  x
         z
                                             "  J         L  t         v                                              F         H  p         r                                              B         D  l         n               !         !  "         $  =%         >%  g&         h&  '         '  (         (  )         )  +         +  9,         :,  c-         d-  .         .  /         /  0         0  2         2  53         63  _4         `4  5         5  6         6  7         7  9         9  1:         2:  [;         \;  <         <  =         =  >         >  @         @  -A         .A  WB         XB  C         C  D         D  E         E  F          G  )H         :H  cI         dI  J         J  K         K  L         L  N         N  5O         6O  _P         `P  Q         Q  R         R  S         S  U         U  1V         2V  [W         \W  X         X  Y         Y  Z         Z  \         \  -]         .]  W^         X^  _         _  `         `  a         a  b          c  )d         *d  Se         Te  }f         ~f  g         g  h          {  |          |  }          ~            N  R                     ʎ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               K!         ^!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 }
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-:-x    :-{I-|    {I-|-\    |-
-    
--P    -{b-    {b-c-	    c-g-D
    g-9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               B!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P           %         &  L         M  s	         t	  
         
           D  j         k                                                                     5            L  0          c  G          z  ~                                     !          F!  J"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            .         0  ^         `                                               N         P  ~	         	  
         
                        >         @  n         p                                    .         0  ^         `                                               N         P  ~                                              >!         @!  n"         %  &         &  ((         *(  X)         Z)  *         *  +         +  ,         ,  .         .  H/         J/  x0         z0  1         1  2         2  4         
4  85         :5  h6         j6  7         7  8         8  9         9  (;         *;  X<         Z<  =         =  >         >  ?         ?  A         A  HB         JB  xC         zC  D         D  E         E  G         
G  8H          P  P          *Q  R          HR  3S          fS  QT          T  oU         h   j         j  0k         2k  `l         bl  m         m  n         n  o         o   q         "q  Pr         Rr  s         s  t         t  u         u  w         w  @x         Bx  py         ry  z         z  {         {   }         }  0~         2~  `         b                                              L         N  {         }             ۈ         ݈             ;                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               :!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 y
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -A-    A-5-x    5-C-|    C-v-\    v--    --h    -b-    b-x-	    x--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               J!         X!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         3  1          f  d                                                          -  	          D	  "
          [
  9         r                                              4                                 +         ,  R         S  y                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                      @  G                                              &  '         '  (         (  
*         *  6+         7+  b,         c,  -         -  .         .  /         /  1         1  >2         ?2  j3         k3  4         4  5         5  6         6  8         8  F9         G9  r:         s:  ;         ;  <         <  =         =  !?         #?  M@         O@  yA         {A  B         B  C         C  D         D  )F         +F  UG         WG  H         H  I         I  J         J  L         L  1M         3M  ]N         _N  O         O  P         P  Q         Q  S         S  9T         ;T  eU         gU  V         V  W         W  X         X  Z         Z  A[         C[  m\         o\  ]         ]  ^         ^  _         _  a         a  Ib         Kb  uc         wc  d         d  e         e  f         f  %h         'h  Qi         Si  }j         j  k         k  l          n  *o         ,o  Vp         Xp  q         q  r         r  s         s  u         u  2v         4v  ^w         `w  x         x  y         y  z         z  |         |  :}         <}  f~         h~                                              B         D  n         p             Ƈ         ȇ                         J         L  v         x             Ύ         Ў             &                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               B!         T!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 }
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-:-x    :-uI-|    uI-|-\    |-
-    
--P    -{b-    {b-c-	    c-a-D
    a-9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ?!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P           %         &  L         M  s	         t	  
         
           ?  e         f                                                                     0            G  ,          ^  C          u  x                                     !          A!  D"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            /         0  _         `                                               O         P  	         	  
         
                        ?         @  o         p                                    /         0  _         `                                               O         P                                                ?!         @!  o"         %  &         &  )(         *(  Y)         Z)  *         *  +         +  ,         ,  .         .  I/         J/  y0         z0  1         1  2         2  	4         
4  95         :5  i6         j6  7         7  8         8  9         9  );         *;  Y<         Z<  =         =  >         >  ?         ?  A         A  IB         JB  yC         zC  D         D  E         E  	G         
G  9H          U  V          V  W          W  X          X  Y          Z  [         h  j         j  1k         2k  al         bl  m         m  n         n  o         o  !q         "q  Qr         Rr  s         s  t         t  u         u  w         w  Ax         Bx  qy         ry  z         z  {         {  }         }  1~         2~  a         b                                              M         N  |         }             ܈         ݈             <                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               :!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 y
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -;-    ;-5-x    5-C-|    C-v-\    v--    --h    -b-    b-x-	    x--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               H!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         3  0          f  c                                                          -  	          D	  #
          [
  :         r                                              4                                  &         '  M         N  t                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         @  G              !          !  "           #  $          @$  G%         &  '         '  (         (  	*         *  5+         7+  a,         c,  -         -  .         .  /         /  1         1  =2         ?2  i3         k3  4         4  5         5  6         6  8         8  E9         G9  q:         s:  ;         ;  <         <  =         =  "?         #?  N@         O@  zA         {A  B         B  C         C  D         D  *F         +F  VG         WG  H         H  I         I  J         J  L         L  2M         3M  ^N         _N  O         O  P         P  Q         Q  S         S  :T         ;T  fU         gU  V         V  W         W  X         X  Z         Z  B[         C[  n\         o\  ]         ]  ^         ^  _         _  a         a  Jb         Kb  vc         wc  d         d  e         e  f         f  &h         'h  Ri         Si  ~j         j  k         k  l          n  +o         ,o  Wp         Xp  q         q  r         r  s         s  u         u  3v         4v  _w         `w  x         x  y         y  z         z  |         |  ;}         <}  g~         h~                                              C         D  o         p             Ǉ         ȇ                         K         L  w         x             ώ         Ў             '                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               @!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 }
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     --    -q;-    q;-J-    J-Y}-l    Y}-!-    !--P    --    -b-h	    b-!-p
    !-/-
    /--    -3׳-L                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               =!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             &         '  M         N  t         u                       r  v                                              >  B          q  ^                                                           E         F  l         m               !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        %  &         &  '         '  )         
)  6*         8*  d+         f+  ,         ,  -         -  .         .  0         0  J1         L1  x2         z2  3         3  4         4  6         6  07         27  ^8         `8  9         9  :         :  ;         ;  =         =  D>         F>  r?         t?  @         @  A         A  B         B  *D         ,D  XE         ZE  F         F  G         G  H         H  J         J  >K         @K  lL         nL  M         M  N         N  O         O  %Q         &Q  SR         TR  S         S  T         T  U         U  W         W  9X         :X  gY         hY  Z         Z  [         [  \         \  ^          ^  M_         N_  {`         |`  a         a  b         b  d         d  3e         4e  af         bf  g         g  h         h  i         i  k         k  Gl         Hl  um         vm  n         n  o         o  p          q  ,r         .r  Zs         \s  t         t  u         u  v         v  x         x  @y         By  nz         pz  {         {  |         |  }         }  &         (  T         V                        ރ            ۄ            	          <  7          j  e                     Ɖ                       "  N         P  |         ~                                b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               G!         ^!  1         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    n
         o
             L                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     P
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-e;-x    e;-I-|    I-M}-\    M}-,-    ,--h    -ob-    ob-߂-	    ߂--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               <!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 #	         $	  J
         K
  q                                                      !  G          H  J                                              $  &          [  H          ~  k                          !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            -         .  [         \                                              A         B  o	         p	  
         
                        '         (  U         V                                              ;         <  i         j                                   !         "  O         P  }         ~                        !         !  5"         6"  c#         %  &         &  '         '  %)         &)  S*         T*  +         +  ,         ,  -         -  /         /  90         :0  g1         h1  2         2  3         3  4         4  6          6  M7         N7  {8         |8  9         9  :         :  <         <  3=         4=  a>         b>  ?         ?  @         @  A         A  C         C  GD         HD  uE         vE  F         F  G         G  H         TI  J         J  K          nL  fM          M  N          N  O          O  P          &Q  R           ?         A  m         o             ɐ         ː             %         '  S         U                        ݗ         ߗ             9         ;  g         i             Ý         ŝ                      !  M         O  {         }             פ         ٤             3         5  a         c                                                      >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               5!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     --    -q;-    q;-J-    J-Y}-l    Y}--    --P    --    -b-h	    b-!-p
    !-/-
    /--    -3׳-L                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ;!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             &         '  M         N  t         u                       r  u                                              >  A          q  _                                                           E         F  l         m               !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        %  &         &  '         '  	)         
)  7*         8*  e+         f+  ,         ,  -         -  .         .  0         0  K1         L1  y2         z2  3         3  4         4  6         6  17         27  _8         `8  9         9  :         :  ;         ;  =         =  E>         F>  s?         t?  @         @  A         A  B         B  +D         ,D  YE         ZE  F         F  G         G  H         H  J         J  ?K         @K  mL         nL  M         M  N         N  O         O  $Q         &Q  RR         TR  S         S  T         T  U         U  
W         W  8X         :X  fY         hY  Z         Z  [         [  \         \  ^          ^  L_         N_  z`         |`  a         a  b         b  d         d  2e         4e  `f         bf  g         g  h         h  i         i  k         k  Fl         Hl  tm         vm  n         n  o         o  p          q  -r         .r  [s         \s  t         t  u         u  v         v  x         x  Ay         By  oz         pz  {         {  |         |  }         }  '         (  U         V                        ߃                      ;         <  i         j             ŉ          Ɖ                        "            P  K          ~  y                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               G!         ^!  1         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    n
         o
             L                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     Q
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-e;-x    e;-I-|    I-M}-\    M}-,-    ,--h    -ob-    ob-߂-	    ߂--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               >!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 #	         $	  J
         K
  q                                                      !  G          H  I                                              $  %          [  I          ~  l                          !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ,         .  Z         \                                              @         B  n	         p	  
         
                        &         (  T         V                                              :         <  h         j                                             "  N         P  |         ~                        !         !  4"         6"  b#         %  &         &  '         '  $)         &)  R*         T*  +         +  ,         ,  -         -  
/         /  80         :0  f1         h1  2         2  3         3  4         4  6          6  L7         N7  z8         |8  9         9  :         :  <         <  2=         4=  `>         b>  ?         ?  @         @  A         A  C         C  FD         HD  tE         vE  F         F  G         G  H         TI  J         J  K          TR  LS          S  zT          T  U          U  V          W  X           @         A  n         o             ʐ         ː             &         '  T         U                        ޗ         ߗ             :         ;  h         i             ĝ         ŝ                       !  N         O  |         }             ؤ         ٤             4         5  b         c                                                      >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               5!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     --    -5-    5-D-    D-w-l    w-4-    4--x    -ub-	    ub--
    -)-
    )-y-|    y-ճ-    ճ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ;!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         #            F  3          i  V            y                                    	          T	  P
          
                                   6         7  ]         ^                                              3         4  Z                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |        D%  n&          p&  }'          '  (          (  *          9*  F+          |+  ,         JL  uM         vM  N         N  O         O  P         P  %R         &R  QS         RS  }T         ~T  U         U  V         V  X         X  -Y         .Y  YZ         ZZ  [         [  \         \  ]         ]  	_         
_  5`         6`  aa         ba  b         b  c         c  d         d  f         f  =g         >g  ih         jh  i         i  j         j  k         k  m         m  En         Fn  qo         ro  p         p  q         q  r         r  !t         "t  Mu         Nu  yv         zv  w         w  x         x  y         y  ){         *{  U|         V|  }         }  ~         ~                        1         2  ]         ^                                              9         :  e         f                                              A         B  m         n             ē         Ɠ                        H         J  t         v             ̚         Κ             $         &  P         R  |         ~             ԡ         ֡              ,         .  X         Z                        ܨ         ި           
  4         6  `         b                                              <         >  h         j                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               <!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    n
         o
             L      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    n
         o
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            0     -,-X     ,--     -/-    /-q;-x    q;-J-|    J-Y}-\    Y}-{-    {--h    -Kb-    Kb-߂-	    ߂--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               B!         V!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             &         '  M         N  t         u                                                           +            N  ;          q  o                                  
            =  ;           E         F  l         m               !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            .         0  ^         `                                               N         P  ~	         	  
         
                        >         @  n         p                                    .         0  ^         `                                               N         P  ~                                              >!         @!  n"         p"  #         D%  s&         t&  '         '  (         (  *         *  3+         4+  c,         d,  -         -  .         .  /         /  #1         $1  S2         T2  3         3  4         4  5         5  7         7  C8         D8  s9         t9  :         :  ;         ;  =         =  3>         4>  c?         d?  @         @  A         A  B         B  #D         $D  SE         TE  F         F  G         G  H          PU  XV          V  W          W  X          
Y  Z          HZ  P[         ir  s         s  t         t  u         u  'w         )w  Wx         Yx  y         y  z         z  {         {  }         }  G~         I~  w         y             ׁ         ف           	  7         9  g         i             Ǉ         ɇ             '         )  W         Y                                              G         I  w         y             ה         ٔ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ,!         6!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,-y-     y--    -5-    5-D-    D-w-l    w-4-    4--x    -ob-	    ob--
    -)-
    )-y-|    y-ճ-    ճ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               9!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         #            F  4          i  W            z                                    	          T	  O
          
                                   6         7  ]         ^                                              3         4  Z                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |        D%  o&          ,  -          .  /          E/  R0          0  1          1  2         JL  tM         vM  N         N  O         O  P         P  $R         &R  PS         RS  |T         ~T  U         U  V         V   X         X  ,Y         .Y  XZ         ZZ  [         [  \         \  ]         ]  _         
_  4`         6`  `a         ba  b         b  c         c  d         d  f         f  <g         >g  hh         jh  i         i  j         j  k         k  m         m  Dn         Fn  po         ro  p         p  q         q  r         r   t         "t  Lu         Nu  xv         zv  w         w  x         x  y         y  ({         *{  T|         V|  }         }  ~         ~                        0         2  \         ^                                              8         :  d         f                                              @         B  l         n             œ         Ɠ                        I         J  u         v             ͚         Κ             %         &  Q         R  }         ~             ա         ֡             -         .  Y         Z                        ݨ         ި  	         
  5         6  a         b                                              =         >  i         j                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               :!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    n
         o
             L      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    n
         o
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            0     -,-X     ,--     -)-    )-q;-x    q;-J-|    J-Y}-\    Y}-{-    {--h    -Qb-    Qb-߂-	    ߂--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               @!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             &         '  M         N  t         u                                                           +            N  <          q  n                                  
            =  :           E         F  l         m               !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            /         0  _         `                                               O         P  	         	  
         
                        ?         @  o         p                                    /         0  _         `                                               O         P                                                ?!         @!  o"         p"  #         D%  r&         t&  '         '  (         (  *         *  2+         4+  b,         d,  -         -  .         .  /         /  "1         $1  R2         T2  3         3  4         4  5         5  7         7  B8         D8  r9         t9  :         :  ;         ;  =         =  2>         4>  b?         d?  @         @  A         A  B         B  "D         $D  RE         TE  F         F  G         G  H          [  \          \  ]          ^  
_          @_  H`          ~`  a         ir  s         s  t         t  u         u  (w         )w  Xx         Yx  y         y  z         z  {         {  }         }  H~         I~  x         y             ؁         ف           	  8         9  h         i             ȇ         ɇ             (         )  X         Y                                              H         I  x         y             ؔ         ٔ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               .!         :!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     --    -k;-x    k;-J-|    J-S}-\    S}-?6-    ?6--h    -b-    b-m-	    m--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               <!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             &         '  M         N  t         u                       r  v                                              >  B          q  ^                                                           E         F  l         m               !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            (         *  R         T  |         ~                                   $         &  N	         P	  x
         z
                                             "  J         L  t         v                                              F         H  p         r                                              B         D  l         n               !         !  "          /  0          1  1          42  03          f3  b4          4  5         K  L         L  N         N  1O         2O  [P         \P  Q         Q  R         R  S         S  U         U  -V         .V  WW         XW  X         X  Y         Y  Z         Z  [          \  )]         *]  S^         T^  }_         ~_  `         `  a         a  b         b  %d         &d  Oe         Pe  yf         zf  g         g  h         h  i         i  !k         "k  Kl         Ll  um         vm  n         n  o           6         8  `         b                        ޚ                    
  2         4  \         ^                        ڡ         ܡ             .         0  X         Z                        ֨         ب              *         ,  T         V  ~                      ү         ԯ             &         (  P         R  z         |             ζ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               U!         V!   2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     +
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-_;-x    _;-I-|    I-G}-\    G}-1-    1--P    -b-    b-c-	    c-7-D
    7--׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               O!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 #	         $	  J
         K
  q                                                      !  G          H  J                                              $  &          [  H          ~  k                          !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            1         2  c         d                                   +         ,  ]         ^  	         	  
         
             %         &  W         X                                               Q         R                                              K         L  }         ~                                   E          F   w!         x!  "         "  #         #  %         %  ?&         @&  q'         r'  (         (  )         )  +         +  9,         :,  k-         l-  .         .  /         /  1         1  32         42  e3         f3  4         4  5         5  6         6  -8         .8  _9         `9  :         :  ;         ;  <         <  '>         (>  Y?         Z?  @         @  A         A  B         B  !D         "D  SE         TE  F         F  G         G  H         H  J         ir  s         s  t         t  u         u  /w         1w  ax         cx  y         y  z         z  {         {  )}         +}  [~         ]~                                   #         %  U         W                                              O         Q                                              I          |              ̭                      E  R                               >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               S!         T!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 m
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-k;-x    k;-J-|    J-S}-\    S}-?6-    ?6--h    -b-    b-m-	    m--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               >!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             &         '  M         N  t         u                       r  u                                              >  A          q  _                                                           E         F  l         m               !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            )         *  S         T  }         ~                                   %         &  O	         P	  y
         z
                                   !         "  K         L  u         v                                              G         H  q         r                                              C         D  m         n               !         !  "          5  6          6  7          .8  *9          `9  \:          :  ;         K  L         L  N         N  0O         2O  ZP         \P  Q         Q  R         R  S         S  U         U  ,V         .V  VW         XW  X         X  Y         Y  Z         Z  [          \  (]         *]  R^         T^  |_         ~_  `         `  a         a  b         b  $d         &d  Ne         Pe  xf         zf  g         g  h         h  i         i   k         "k  Jl         Ll  tm         vm  n         n  o           7         8  a         b                        ߚ           	         
  3         4  ]         ^                        ۡ         ܡ             /         0  Y         Z                        ר         ب             +         ,  U         V                        ӯ         ԯ             '         (  Q         R  {         |             ϶                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               U!         V!   2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     *
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-_;-x    _;-I-|    I-G}-\    G}-S7-    S7--P    -b-    b-c-	    c-=-D
    =--׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               L!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 #	         $	  J
         K
  q                                                      !  G          H  I                                              $  %          [  I          ~  l                          !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            0         2  b         d                                   *         ,  \         ^  	         	  
         
             $         &  V         X                                               P         R                                              J         L  |         ~                                   D          F   v!         x!  "         "  #         #  %         %  >&         @&  p'         r'  (         (  )         )  +         +  8,         :,  j-         l-  .         .  /         /   1         1  22         42  d3         f3  4         4  5         5  6         6  ,8         .8  ^9         `9  :         :  ;         ;  <         <  &>         (>  X?         Z?  @         @  A         A  B         B   D         "D  RE         TE  F         F  G         G  H         H  J         ir  s         s  t         t  u         u  0w         1w  bx         cx  y         y  z         z  {         {  *}         +}  \~         ]~                                   $         %  V         W                                              P         Q                                              J          ˱  ز                      Q  ^                      ׶                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               S!         T!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 m
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-=-     =--    -5-x    5-D-|    D-w-\    w-1-    1--h    -ub-    ub-m-	    m--h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     \            !          L!  1          1  ~=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         !  
          B  +          c  L            m                                    	          P	  T
          
                                   <         =  e         f                                   &         '  O         P  x                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            4         6  j         l                                   B         D  x         z  	         	  
         
             P         R                                   (         *  ^         `                                    6         8  l         n                                   D         F  z         |               !         !  #         #  R$          <  =          (>  (?          `?  `@          @  A          A  B          I  UJ         VJ  K         K  L         L  M         M  -O         .O  cP         dP  Q         Q  R         R  T         T  ;U         <U  qV         rV  W         W  X         X  Z         Z  I[         J[  \         \  ]         ]  ^         ^  !`         "`  Wa         Xa  b         b  c         c  d         d  /f         0f  eg         fg  h         h  i         i  k         k  =l         >l  sm         tm  n         n  o         o  q         q  Kr         Lr  s         s  t         t  u         u  #w         $w  Yx         Zx  y         y  z         z  {         {  1}         2}  g~         h~             Ӏ         Ԁ  	         
  ?         @  u         v                                   M         N                                   %         &  [         \             Ǒ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               K!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     +
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,-I-     I--    -S;-x    S;-I-|    I-;}-\    ;}--    --h    -{b-    {b-͂-	    ͂--h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     g            !          N!  1          1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             (         )  Q         R  z         {                                                           %            F  /          g  i                                              +  -           )         *  R         S  {          |   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            6         8  n         p                                   N         P             	         	  
         
  .         0  f         h                                   F         H  ~                                 &         (  ^         `                                   >         @  v         x                          "          "  V#         X#  $         $  %         %  &          '  7(         8(  o)         p)  *         *  +         +  -         -  O.         P.  /         /  0         0  1         1  /3         03  g4         h4  5         5  6         6  8         8  G9         H9  :         :  ;         ;  <         <  '>         (>  _?         `?  @         @  A         A  C         C  ?D         @D  wE         xE  F         F  G         G  I          Y  Z          WZ  b[          [  \          \  ]           ^  +_         o   q         q  8r         :r  ps         rs  t         t  u         u  w         w  Px         Rx  y         y  z         z  {         {  0}         2}  h~         j~             ؀         ڀ             H         J                                   (         *  `         b             Ћ         ҋ           
  @         B  x         z                                  "  X                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               M!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,-=-     =--    -5-x    5-D-|    D-w-\    w-+-    +--h    -ub-    ub-m-	    m--h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     ]            !          L!  1          1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         !            B  ,          c  M            n                                    	          P	  S
          
                                   <         =  e         f                                   &         '  O         P  x                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            5         6  k         l                                   C         D  y         z  	         	  
         
             Q         R                                   )         *  _         `                                   7         8  m         n                                   E         F  {         |               !         !  #         #  S$          C  D          @D  @E          xE  xF          F  G          G  H          I  TJ         VJ  K         K  L         L  M         M  ,O         .O  bP         dP  Q         Q  R         R  T         T  :U         <U  pV         rV  W         W  X         X  Z         Z  H[         J[  ~\         \  ]         ]  ^         ^   `         "`  Va         Xa  b         b  c         c  d         d  .f         0f  dg         fg  h         h  i         i  k         k  <l         >l  rm         tm  n         n  o         o  q         q  Jr         Lr  s         s  t         t  u         u  "w         $w  Xx         Zx  y         y  z         z  {         {  0}         2}  f~         h~             Ҁ         Ԁ           
  >         @  t         v                                   L         N                                   $         &  Z         \             Ƒ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               K!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     *
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,-O-     O--    -S;-x    S;-I-|    I-;}-\    ;}--    --h    -{b-    {b-͂-	    ͂--h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     h            !          R!  1          1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             (         )  Q         R  z         {                                                           %            F  0          g  h                                              +  ,           )         *  R         S  {          |   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            7         8  o         p                                   O         P             	         	  
         
  /         0  g         h                                   G         H                                   '         (  _         `                                   ?         @  w         x                          "          "  W#         X#  $         $  %         %  &          '  6(         8(  n)         p)  *         *  +         +  -         -  N.         P.  /         /  0         0  1         1  .3         03  f4         h4  5         5  6         6  8         8  F9         H9  ~:         :  ;         ;  <         <  &>         (>  ^?         `?  @         @  A         A  C         C  >D         @D  vE         xE  F         F  G         G  I          c_  n`          `  a          a  b          ,c  7d          od  ze         o  q         q  9r         :r  qs         rs  t         t  u         u  w         w  Qx         Rx  y         y  z         z  {         {  1}         2}  i~         j~             ـ         ڀ             I         J                                   )         *  a         b             ы         ҋ  	         
  A         B  y         z                        !         "  Y                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               M!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,-I-     I--    -:-x    :-iI-|    iI-|-\    |--    --h    -b-    b-_x-	    _x--h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     ^            !          P!  1          1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      F         G  o	         p	  
         
           ?  g         h                                                         %            >            W  3          p  L                                                 !          M!  E"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            2         4  f         h                                   6         8  j         l  	         	  
         
             :         <  n         p                        
           >         @  r         t                                   B         D  v         x                                   F         H  z          |   !         !  "         "  $         $  J%         L%  ~&         &  '         '  (         (  *         *  N+         P+  ,         ,  -         -  .         .  0          0  R1         T1  2         2  3         3  4         4  "6         $6  V7         X7  8         8  9         9  :         :  &<         (<  Z=         \=  >         >  ?         ?  @         @  *B         ,B  ^C         `C  D         D  E         E  F         F  .H         ^H  I         I  J         J  K         K  -M         .M  aN         bN  O         O  P         P  Q         Q  1S         2S  eT         fT  U         U  V         V  X         X  5Y         6Y  iZ         jZ  [         [  \         \  ^         ^  9_         :_  m`         n`  a         a  b         b  	d         
d  =e         >e  qf         rf  g         g  h         h  j         j  Ak         Bk  ul          tm  mn          n  o          o  p          q  r          Lr  Es                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               L!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 g
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            4     -,-`     ,-O-     O--    -k;-    k;-J-    J-S}-d    S}-7-    7--0    --    -b-H	    b-c-P
    c-+-
    +--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     g            !          P!  1          1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 	          	  H
         I
  q                                 	         
  2         3  [          \  V                                              @  :          y  b                                      !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q   w  operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       \            9         :  s         t                        !         "  [         \             	         	  	         
  C         D  }         ~                        +         ,  e         f                                   M         N                                   5         6  o         p                        !         !  W"         X"  #         #  $         $  &         &  ?'         @'  y(         z(  )         )  *         *  ',         (,  a-         b-  .         .  /         /  1         1  I2         J2  3         3  4         4  5         5  17         27  k8         l8  9         9  :         :  <         <  S=         T=  >         >  ?         ?  A         A  ;B         <B  uC         vC  D         D  E         E  #G         $G  \H         vl  m         m  n         n  "p         $p  \q         ^q  r         r  s         s  
u         u  Dv         Fv  ~w         w  x         x  y         y  ,{         .{  f|         h|  }         }  ~         ~             N         P                       ă  Ʉ                      8  =          r  w                                   X         Z             ̎                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               b!         l!  2         "2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 k
                     
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,-C-     C--    -:-x    :-cI-|    cI-|-\    |--    --h    -b-    b-_x-	    _x--h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     ]            !          L!  1          1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      F         G  o	         p	  
         
           ?  g         h                                                         %            >            W  4          p  M                                                 !          M!  D"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            3         4  g         h                                   7         8  k         l  	         	  
         
             ;         <  o         p                                   ?         @  s         t                                   C         D  w         x                                   G         H  {          |   !         !  "         "  $         $  K%         L%  &         &  '         '  (         (  *         *  O+         P+  ,         ,  -         -  .         .  0          0  S1         T1  2         2  3         3  4         4  #6         $6  W7         X7  8         8  9         9  :         :  '<         (<  [=         \=  >         >  ?         ?  @         @  +B         ,B  _C         `C  D         D  E         E  F         F  /H         ^H  I         I  J         J  K         K  ,M         .M  `N         bN  O         O  P         P  Q         Q  0S         2S  dT         fT  U         U  V         V   X         X  4Y         6Y  hZ         jZ  [         [  \         \  ^         ^  8_         :_  l`         n`  a         a  b         b  d         
d  <e         >e  pf         rf  g         g  h         h  j         j  @k         Bk  tl          s  {t          t  u          u  v          $w  x          Zx  Sy                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               N!         R!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 g
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            4     -' -`     ' -հ-     հ-۰-    ۰-i4-x    i4-'@-\    '@-h-<    h--    -1-    1--    -b- 	    b-c-(
    c-+-
    +--
                   =         b      J      7   U   7   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     h            !          R!  1                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 64 })  outputElements  5184    input0Elements  31104   input1Elements  24576   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     4                           #         b      J      7   M     operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       @                                            '	           4         5  ]         ^                                               "            [  T            ~                                                                  b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     4                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  1728    input0Elements  1728    invocations     1                           3         b      J      7   Q   w  operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       \            8         :  r         t                                  "  Z         \             	         	           
  B         D  |         ~                        *         ,  d         f                                   L         N                                   4         6  n         p                        !         !  V"         X"  #         #  $         $  &         &  >'         @'  x(         z(  )         )  *         *  &,         (,  `-         b-  .         .  /         /  1         1  H2         J2  3         3  4         4  5         5  07         27  j8         l8  9         9  :         :  <         <  R=         T=  >         >  ?         ?   A         A  :B         <B  tC         vC  D         D  E         E  "G         $G  ]H         vl  m         m  n         n  #p         $p  ]q         ^q  r         r  s         s  u         u  Ev         Fv  w         w  x         x  y         y  -{         .{  g|         h|  }         }  ~         ~             O         P             Ã         ă             7         8  q         r                                      %          Z  _                                          b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     4                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               `!         h!  2         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 k
                     
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,--     --    -6-x    6-D-|    D-w-\    w--    --h    -3b-    3b-ex-	    ex--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ;!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         !  
          B  +          c  L            m                                    	          P	  T
          
                                   <         =  e         f                                   &         '  O         P  x                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       
          @  :          p  j                     $  %         %  '         '  8(         :(  j)         l)  *         *  +         +   -         -  2.         4.  d/         f/  0         0  1         1  2         2  ,4         .4  ^5         `5  6         6  7         7  8         8  &:         (:  X;         Z;  <         <  =         =  >         >   @         "@  RA         TA  B         B  C         C  D         D  F         F  LG         NG  ~H         H  I         I  K         K  BL         DL  tM         vM  N         N  O         O  
Q         Q  <R         >R  nS         pS  T         T  U         U  W         W  7X         8X  iY         jY  Z         Z  [         [  \          ]  1^         2^  c_         d_  `         `  a         a  b         b  +d         ,d  ]e         ^e  f         f  g         g  h         h  %j         &j  Wk         Xk  l         l  m         m  n         n  p          p  Qq         Rq  r         r  s         s  t         t  v         v  Kw         Lw  }x         ~x  y         y  z         z  |         |  E}         F}  w~         x~             ۀ         ܀             ?         @  q         r             Ն         ֆ             9         :  k         l             ό         Ќ             3         d                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     \             !          H!  1          1  x=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 g
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-Y;-x    Y;-I-|    I-A}-\    A}-.-    .--h    -cb-    cb-n-	    n--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               D!         X!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             (         )  Q         R  z         {                                                           %            F  /          g  i                                              +  -           )         *  R         S  {          |   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            (         *  R         T  |         ~                                   $         &  N	         P	  x
         z
                                             "  J         L  t         v                                              F         H  p         r                                              B         D  l         n               !          !  "          "  #          $  $          +%  &          N&  <'         @G  iH         jH  I         I  J         J  K         K  "M         $M  LN         NN  vO         xO  P         P  Q         Q  R         R  T          T  HU         JU  rV         tV  W         W  X         X  Y         Y  [         [  D\         F\  n]         p]  ^         ^  _         _  `         `  b         b  @c         Bc  jd         ld  e         e  f         f  g         g  i         i  <j         >j  fk         hk  l         l  m           ͏         Ώ             !         "  K         L  u         v             ɖ         ʖ                        G         H  q         r             ŝ         Ɲ                        C         D  m         n                      ¤                        ?         @  i         j                                              ;                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     g            !          b!  1          2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     O
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-y-     y--    -6-x    6-D-|    D-w-\    w--    --h    -3b-    3b-ex-	    ex--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               9!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         !            B  ,          c  M            n                                    	          P	  S
          
                                   <         =  e         f                                   &         '  O         P  x                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                  0  *          `  Z                     $  %         %  '         '  9(         :(  k)         l)  *         *  +         +  -         -  3.         4.  e/         f/  0         0  1         1  2         2  -4         .4  _5         `5  6         6  7         7  8         8  ':         (:  Y;         Z;  <         <  =         =  >         >  !@         "@  SA         TA  B         B  C         C  D         D  F         F  MG         NG  H         H  I         I  K         K  CL         DL  uM         vM  N         N  O         O  Q         Q  =R         >R  oS         pS  T         T  U         U  W         W  6X         8X  hY         jY  Z         Z  [         [  \          ]  0^         2^  b_         d_  `         `  a         a  b         b  *d         ,d  \e         ^e  f         f  g         g  h         h  $j         &j  Vk         Xk  l         l  m         m  n         n  p          p  Pq         Rq  r         r  s         s  t         t  v         v  Jw         Lw  |x         ~x  y         y  z         z  |         |  D}         F}  v~         x~             ڀ         ܀             >         @  p         r             Ԇ         ֆ             8         :  j         l             Ό         Ќ              2         d                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     ]            !          H!  1          1  y=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 g
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-Y;-x    Y;-I-|    I-A}-\    A}-.-    .--h    -ib-    ib-n-	    n--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               D!         X!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             (         )  Q         R  z         {                                                           %            F  0          g  h                                              +  ,           )         *  R         S  {          |   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            )         *  S         T  }         ~                                   %         &  O	         P	  y
         z
                                   !         "  K         L  u         v                                              G         H  q         r                                              C         D  m         n               !          q'  _(          (  )          )  *          *  +          +  ,         @G  hH         jH  I         I  J         J  K         K  #M         $M  MN         NN  wO         xO  P         P  Q         Q  R         R  T          T  IU         JU  sV         tV  W         W  X         X  Y         Y  [         [  E\         F\  o]         p]  ^         ^  _         _  `         `  b         b  Ac         Bc  kd         ld  e         e  f         f  g         g  i         i  =j         >j  gk         hk  l         l  m           ̏         Ώ                       "  J         L  t         v             Ȗ         ʖ                        F         H  p         r             ĝ         Ɲ                        B         D  l         n                      ¤                        >         @  h         j                                              :                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     h            !          f!  1          2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     N
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-g-     g--    -:-x    :-oI-|    oI-|-\    |-1-    1--h    -9b-    9b--	    --h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               /!         @!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      F         G  o	         p	  
         
           ?  g         h                                                         %            >            W  3          p  L                                                 !          M!  E"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            .         0  ^         `                                               N         P  ~	         	  
         
                        >         @  n         p                                    .         0  ^         `                                               N         P  ~                                              >!         @!  n"         p"  #         #  $         $  %          &  .'         0'  ^(         `(  )         )  *         *  +         +  -          -  N.         P.  ~/         /  0         0  1         1  3         3  >4         @4  n5         p5  6         6  7         7  8          9  .:         0:  ^;         `;  <         <  =         =  >         >  @          @  NA         PA  ~B         B  C         C  D         D  F         F  >G         @G  nH         pH  I         I  J          d  e          e  f          &g  (h          bh  di          i  j         d                      Ĕ             "         $  R         T                                              B         D  r         t             ҡ         ԡ             2         4  b         d             §         ħ             "         $  R         T                                              B         D  r                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     Y                       L!  1          1  {=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   2      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-5-x    5-C-|    C-v-\    v-	-    	--h    -{b-    {b-!n-	    !n--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               @!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        1  3          b  d                                                          '  	          @	  &
          Y
  ?         r                                              >                                 6         7  _         `                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            *         ,  V         X                                              2         4  ^	         `	  
         
                                   :         <  f         h                                              B         D  n         p                                               J         L  v         x                          !         !  &#          0/  .0          ^0  \1          1  2          2  3          3  4         HI  rJ         tJ  K         K  L         L  M         M  "O         $O  NP         PP  zQ         |Q  R         R  S         S  T          U  *V         ,V  VW         XW  X         X  Y         Y  Z         Z  \         \  2]         4]  ^^         `^  _         _  `         `  a         a  c         c  :d         <d  fe         he  f         f  g         g  h         h  j         j  Bk         Dk  nl         pl  m         m  n         n  o         o  q          q  Jr         Lr  vs         xs  t         t  u         u  v         v  &x         (x  Ry         Ty  ~z         z  {         {  |         |  ~         ~  .         0  Z         \                        ރ           
           6         8  b         d                                              >         @  j         l                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               @!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     O
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-s-     s--    -:-x    :-oI-|    oI-|-\    |-1-    1--h    -3b-    3b--	    --h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               3!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      F         G  o	         p	  
         
           ?  g         h                                                         %            >            W  4          p  M                                                 !          M!  D"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            /         0  _         `                                               O         P  	         	  
         
                        ?         @  o         p                                    /         0  _         `                                               O         P                                                ?!         @!  o"         p"  #         #  $         $  %          &  /'         0'  _(         `(  )         )  *         *  +         +  -          -  O.         P.  /         /  0         0  1         1  3         3  ?4         @4  o5         p5  6         6  7         7  8          9  /:         0:  _;         `;  <         <  =         =  >         >  @          @  OA         PA  B         B  C         C  D         D  F         F  ?G         @G  oH         pH  I         I  J          j  k          l  m          Rm  Tn          n  o          o  p         d             Ô         Ĕ             #         $  S         T                                              C         D  s         t             ӡ         ԡ             3         4  c         d             ç         ħ             #         $  S         T                                              C         D  s                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     X                       H!  1          1  t=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   1      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-5-x    5-C-|    C-v-\    v-	-    	--h    -{b-    {b-!n-	    !n--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               @!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        1  2          b  c                                                          '  	          @	  '
          Y
  @         r                                              >                                 6         7  _         `                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            +         ,  W         X                                              3         4  _	         `	  
         
                                   ;         <  g         h                                              C         D  o         p                                               K         L  w         x                          !         !  '#          5  6          D6  B7          r7  p8          8  9          9  :         HI  sJ         tJ  K         K  L         L  M         M  #O         $O  OP         PP  {Q         |Q  R         R  S         S  T          U  +V         ,V  WW         XW  X         X  Y         Y  Z         Z  \         \  3]         4]  _^         `^  _         _  `         `  a         a  c         c  ;d         <d  ge         he  f         f  g         g  h         h  j         j  Ck         Dk  ol         pl  m         m  n         n  o         o  q          q  Kr         Lr  ws         xs  t         t  u         u  v         v  'x         (x  Sy         Ty  z         z  {         {  |         |  ~         ~  /         0  [         \                        ߃                      7         8  c         d                                              ?         @  k         l                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               @!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     N
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -;-    ;-_;-x    _;-I-|    I-G}-\    G}--    --h    -Wb-    Wb-	-	    	--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               G!         V!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             (         )  Q         R  z         {                       r  j                                              6  .          g  P            q                                               )         *  R         S  {          |   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ,         .  Z         \                                              @         B  n	         p	  
         
                        &         (  T         V                                              :         <  h         j                                             "  N         P  |         ~                        !         !  4"         6"  b#         d#  $         $  %         %  &         &  (         (  H)         J)  v*         x*  +         +  ,         ,   .         .  ./         0/  \0         ^0  1         1  2         2  3         3  5         5  B6         D6  p7         r7  8         8  9         9  :         :  (<         *<  V=         X=  >         >  ?         ?  @         @  B         B  <C         >C  jD         lD  E         E  F          F  G          G  H          
I  I          +J  K          LK  3L         l  m         m  o          o  Mp         Np  {q         |q  r         r  s         s  u         u  3v         4v  aw         bw  x         x  y         y  z         z  |         |  G}         H}  u~         v~             р         Ҁ              -         .  [         \                                              A         B  o         p             ˍ         ̍             '                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               +!         :!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   2      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-w;-x    w;-J-|    J-_}-\    _}-m-    m--h    -b-    b-x-	    x--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               D!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 	          	  H
         I
  q                                 	         
  2         3  [          \  V                                              @  :          y  b                                      !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            -         .  [         \                                              A         B  o	         p	  
         
                        '         (  U         V                                              ;         <  i         j                                   !         "  O         P  }         ~                        !         !  5"         6"  c#         d#  $         $  %         %  &         &  (         (  H)         J)  v*         x*  +         +  ,         ,   .         .  ./         0/  \0         ^0  1         1  2         2  3         3  5         5  B6         D6  p7         r7  8         8  9         9  :         :  (<         *<  V=         X=  >         >  ?         ?  @         @  B         B  <C         >C  jD         lD  E         E  F         F  G         G  "I         $I  PJ         RJ  ~K         K  L         L  M         M  O         
O  6P         8P  dQ         fQ  R         R  S         S  T         T  V         V  JW         LW  xX         zX  Y         Y  Z         Z  \         \  0]         2]  ^^         `^  _         _  `         `  a         a  c         c  Dd         Fd  re         te  f         f  g         g  h                      چ  ؇                      V  T                     ʐ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               C!         V!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 {
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -;-    ;-_;-x    _;-I-|    I-G}-\    G}--    --h    -Wb-    Wb-	-	    	--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               G!         V!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             (         )  Q         R  z         {                       r  i                                              6  -          g  Q            r                                               )         *  R         S  {          |   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            -         .  [         \                                              A         B  o	         p	  
         
                        '         (  U         V                                              ;         <  i         j                                   !         "  O         P  }         ~                        !         !  5"         6"  c#         d#  $         $  %         %  &         &  (         (  I)         J)  w*         x*  +         +  ,         ,  .         .  //         0/  ]0         ^0  1         1  2         2  3         3  5         5  C6         D6  q7         r7  8         8  9         9  :         :  )<         *<  W=         X=  >         >  ?         ?  @         @  B         B  =C         >C  kD         lD  E         E  F          mL  TM          M  uN          N  O          O  P          P  Q         l  m         m  o          o  Lp         Np  zq         |q  r         r  s         s  u         u  2v         4v  `w         bw  x         x  y         y  z         z  |         |  F}         H}  t~         v~             Ѐ         Ҁ              ,         .  Z         \                                              @         B  n         p             ʍ         ̍             &                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               +!         :!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   1      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-};-x    };-J-|    J-e}-\    e}-{-    {--h    -b-    b-x-	    x--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               @!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 	          	  H
         I
  q                                 	         
  2         3  [          \  U                                               @  9          y  c                                      !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ,         .  Z         \                                              @         B  n	         p	  
         
                        &         (  T         V                                              :         <  h         j                                             "  N         P  |         ~                        !         !  4"         6"  b#         d#  $         $  %         %  &         &  (         (  I)         J)  w*         x*  +         +  ,         ,  .         .  //         0/  ]0         ^0  1         1  2         2  3         3  5         5  C6         D6  q7         r7  8         8  9         9  :         :  )<         *<  W=         X=  >         >  ?         ?  @         @  B         B  =C         >C  kD         lD  E         E  F         F  G         G  #I         $I  QJ         RJ  K         K  L         L  M         M  	O         
O  7P         8P  eQ         fQ  R         R  S         S  T         T  V         V  KW         LW  yX         zX  Y         Y  Z         Z  \         \  1]         2]  _^         `^  _         _  `         `  a         a  c         c  Ed         Fd  se         te  f         f  g         g  h          ҋ  Ќ                      N  L                     ʐ                                 >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               E!         Z!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 {
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -A-    A-:-    :-uI-    uI-|-d    |-q
-    q
--0    --    -b-H	    b-c-P
    c-O-
    O-9׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               H!         X!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      F         G  o	         p	  
         
           ?  g         h                                                         %            >            W  3          p  L                                                 !          M!  E"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q   w  operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       \            *         ,  V         X                                              2         4  ^	         `	  
         
                                   :         <  f         h                                              B         D  n         p                                               J         L  v         x                          !         !  &#         (#  R$         T$  ~%         %  &         &  '         '  )         )  .*         0*  Z+         \+  ,         ,  -         -  .         .  
0         0  61         81  b2         d2  3         3  4         4  5         5  7         7  >8         @8  j9         l9  :         :  ;         ;  <         <  >         >  F?         H?  r@         t@  A         A  B         B  C         h  (j         *j  Tk         Vk  l         l  m         m  n         n  p         p  0q         2q  \r         ^r  s         s  t          t  u          u  v          w  x          :x  1y          fy  ]z         z  {         {  |         |  ~         ~  @         B  l         n                      Ă                        C         E  o         q             ǉ         ɉ                                b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               @!         X!  1         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 w
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -;-    ;-5-x    5-C-|    C-w-\    w-i0-    i0--h    -b-    b-x-	    x--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               F!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        1  3          b  d                                                          '  	          @	  &
          Y
  ?         r                                              >                                 6         7  _         `                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                        2  1          d  c                                 %  	'         
'  9(         :(  i)         j)  *         *  +         +  ,         ,  ).         *.  Y/         Z/  0         0  1         1  2         2  4         4  I5         J5  y6         z6  7         7  8         8  	:         
:  9;         :;  i<         j<  =         =  >         >  ?         ?  )A         *A  YB         ZB  C         C  D         D  E         E  G         G  IH         JH  yI         h  i         i  *k         ,k  Zl         \l  m         m  n         n  o         o  q         q  Jr         Lr  zs         |s  t         t  u         u  
w         w  :x         <x  jy         ly  z         z  {         {  |         |  *~         ,~  Z         \                                              J         L  z         |             ڈ         ܈  
           :         <  j         P  ~                      ސ                      >         @  n         p             Ζ         Ж              .         0  ^         `                                               N         P  ~                      ޣ                      >         @  n         p             Ω         Щ              .         0  ^         `                                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               >!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 {
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -;-    ;-:-    :-uI-    uI-|-d    |-q
-    q
--0    --    -b-H	    b-c-P
    c-O-
    O-9׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               G!         V!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      F         G  o	         p	  
         
           ?  g         h                                                         %            >            W  4          p  M                                                 !          M!  D"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q   w  operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       \            +         ,  W         X                                              3         4  _	         `	  
         
                                   ;         <  g         h                                              C         D  o         p                                               K         L  w         x                          !         !  '#         (#  S$         T$  %         %  &         &  '         '  )         )  /*         0*  [+         \+  ,         ,  -         -  .         .  0         0  71         81  c2         d2  3         3  4         4  5         5  7         7  ?8         @8  k9         l9  :         :  ;         ;  <         <  >         >  G?         H?  s@         t@  A         A  B         B  C         h  )j         *j  Uk         Vk  l         l  m         m  n         n  p         p  1q         2q  ]r         ^r  s         s  t         t  u         u  w         w  9x         :x  ey         fy  z          z  {          {  |          |  }          ~            B  9         n             Â         Ă                        D         E  p         q             ȉ         ɉ                                b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               >!         T!  1         1  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 w
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-5-x    5-C-|    C-w-\    w-i0-    i0--h    -b-    b-x-	    x--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               B!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                        1  2          b  c                                                          '  	          @	  '
          Y
  @         r                                              >                                 6         7  _         `                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                     ,  +          ^  ]	          	  
          
           %  '         
'  8(         :(  h)         j)  *         *  +         +  ,         ,  (.         *.  X/         Z/  0         0  1         1  2         2  4         4  H5         J5  x6         z6  7         7  8         8  :         
:  8;         :;  h<         j<  =         =  >         >  ?         ?  (A         *A  XB         ZB  C         C  D         D  E         E  G         G  HH         JH  xI         h  i         i  +k         ,k  [l         \l  m         m  n         n  o         o  q         q  Kr         Lr  {s         |s  t         t  u         u  w         w  ;x         <x  ky         ly  z         z  {         {  |         |  +~         ,~  [         \                                              K         L  {         |             ۈ         ܈             ;         <  k         P                        ߐ                      ?         @  o         p             ϖ         Ж              /         0  _         `                                               O         P                        ߣ                      ?         @  o         p             ϩ         Щ              /         0  _         `                                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               >!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 {
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -M-    M-k;-    k;-J-    J-S}-d    S}-G-    G--p    -b-    b-3-
    3-A-
    A--t    -9׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               L!         `!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             (         )  Q         R  z         {                       r  j                                              6  .          g  P            q                                               )         *  R         S  {          |   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                 (  /          d  k                     %  1&         2&  ['         \'  (         (  )         )  *         *  ,         ,  --         .-  W.         X.  /         /  0         0  1         1  2          3  )4         *4  S5         T5  }6         ~6  7         7  8         8  9         9  %;         &;  O<         P<  y=         z=  >         >  ?         ?  @         @  !B         "B  KC         LC  uD         vD  E         E  F         F  G         G  I         I  FJ         HJ  pK         rK  L         L  M         M  N         N  P         P  BQ         DQ  lR         nR  S         S  T         T  U         U  W         W  >X         @X  hY         jY  Z         Z  [         [  \         \  ^         ^  :_         <_  d`         f`  a         a  b         b  c         c  e         e  6f         8f  `g         bg  h         h  i         i  j         j  l         
l  2m         4m  \n         ^n  o         o  p         p  q         q  s         s  .t         0t  Xu         Zu  v         v  w         w  x         x   z         z  *{         ,{  T|         V|  ~}         }  ~         ~                        &         (  P         R  z         |             Ά         І             "         $  L         N  v         x             ʍ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               Q!         j!  2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    p
         q
             R      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     H
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-;-x    ;-J-|    J-k}-\    k}-C-    C--h    -b-    b-傳-	    傳--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               :!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 	          	  H
         I
  q                                 	         
  2         3  [          \  V                                              @  :          y  b                                      !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            1         2  c         d                                   +         ,  ]         ^  	         	  
         
             %         &  W         X                                               Q         R                                              K         L  }         ~                                   E          F   w!         x!  "         "  #          0  1          1  2           3  3          *4  5          T5  F6         G  H         H  I         I  -K         .K  _L         `L  M         M  N         N  O         O  'Q         (Q  YR         ZR  S         S  T         T  U         U  !W         "W  SX         TX  Y         Y  Z         Z  [         [  ]         ]  M^         N^  _         _  `         `  a         a  c         c  Gd         Hd  ye         ze  f         f  g         g  i         i  Aj         Bj  sk         tk  l         l  m         Pp  q         q  r         r  s         s  u         u  Hv         Jv  zw         |w  x         x  y         y  {         {  B|         D|  t}         v}  ~         ~             
           <         >  n         p             ҅         ԅ             6         8  h         j             ̋         ΋              0         2  b         d             Ƒ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               =!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -G-    G-k;-    k;-J-    J-S}-d    S}-M-    M--p    -b-    b-3-
    3-A-
    A--t    -9׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               J!         \!  1         
2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             (         )  Q         R  z         {                       r  i                                              6  -          g  Q            r                                               )         *  R         S  {          |   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                        !          T!  ["          "  #          #  $         %  0&         2&  Z'         \'  (         (  )         )  *         *  ,         ,  ,-         .-  V.         X.  /         /  0         0  1         1  2          3  (4         *4  R5         T5  |6         ~6  7         7  8         8  9         9  $;         &;  N<         P<  x=         z=  >         >  ?         ?  @         @   B         "B  JC         LC  tD         vD  E         E  F         F  G         G  I         I  GJ         HJ  qK         rK  L         L  M         M  N         N  P         P  CQ         DQ  mR         nR  S         S  T         T  U         U  W         W  ?X         @X  iY         jY  Z         Z  [         [  \         \  ^         ^  ;_         <_  e`         f`  a         a  b         b  c         c  e         e  7f         8f  ag         bg  h         h  i         i  j         j  	l         
l  3m         4m  ]n         ^n  o         o  p         p  q         q  s         s  /t         0t  Yu         Zu  v         v  w         w  x         x  z         z  +{         ,{  U|         V|  }         }  ~         ~                        '         (  Q         R  {         |             φ         І             #         $  M         N  w         x             ˍ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               Q!         j!  2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    p
         q
             R      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     I
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-;-x    ;-J-|    J-k}-\    k}-I-    I--h    -{b-    {b-傳-	    傳--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               <!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 	          	  H
         I
  q                                 	         
  2         3  [          \  U                                               @  9          y  c                                      !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            0         2  b         d                                   *         ,  \         ^  	         	  
         
             $         &  V         X                                               P         R                                              J         L  |         ~                                   D          F   v!         x!  "         "  #          ~6  p7          7  8          8  9          9  :          &;  <         G  H         H  I         I  ,K         .K  ^L         `L  M         M  N         N  O         O  &Q         (Q  XR         ZR  S         S  T         T  U         U   W         "W  RX         TX  Y         Y  Z         Z  [         [  ]         ]  L^         N^  ~_         _  `         `  a         a  c         c  Fd         Hd  xe         ze  f         f  g         g  i         i  @j         Bj  rk         tk  l         l  m         Pp  q         q  r         r  s         s  u         u  Iv         Jv  {w         |w  x         x  y         y  {         {  C|         D|  u}         v}  ~         ~                        =         >  o         p             Ӆ         ԅ             7         8  i         j             ͋         ΋              1         2  c         d             Ǒ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               9!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -G-    G-=6-x    =6-D-|    D-%x-\    %x--    --h    -b-    b-n-	    n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               H!         X!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         !  
          B  +          c  L            m                                    	          P	  Q
          
                                   <         =  e         f                                   ;         <  k         l                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            8         :  r         t                                  "  Z         \             	         	           
  B         D  |         ~                        *         ,  d         f                                   L         N                                   4         6  n         p                        !         !  V"         X"  #         #  $          $  %           &  &          4'  0(          h(  d)          )  *         H  J         J  WK         XK  L         L  M         M  O         O  ?P         @P  yQ         zQ  R         R  S         S  'U         (U  aV         bV  W         W  X         X  Z         Z  I[         J[  \         \  ]         ]  ^         ^  1`         2`  ka         la  b         b  c         c  e         e  Sf         Tf  g         g  h         h  j         j  ;k         <k  ul         vl  m         No  p         p  q         q  r         r  5t         6t  ou         pu  v         v  w         w  y         y  Wz         Xz  {         {  |         |  ~         ~  ?         @  y         z                        '         (  a         b             Շ         և             I         J                                   1         2  k         l             ߒ                              >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               H!         X!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     +
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -A-    A-;-x    ;-)J-|    )J-w}-\    w}--    --h    -ob-    ob-傳-	    傳--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               J!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             (         )  Q         R  z         {                                                           %            F  /          g  d                                              +  (         $  L         M  u         v               !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            2         4  f         h                                   6         8  j         l  	         	  
         
             :         <  n         p                        
           >         @  r         t                                   B         D  v         x                                   F         H  z          |   !         !  "         "  $         $  %          &  3'         4'  g(         h(  )         )  *         *  ,         ,  7-         8-  k.         l.  /         /  0         0  2         2  ;3         <3  o4         p4  5         5  6         6  8         8  ?9         @9  s:         t:  ;         ;  <         <  >         >  C?         D?  w@         x@  A         A  B         B  D         D  GE         HE  {F         |F  G         G  H          la  nb          b  c          c  d          e  f          Tf  Vg         "o  Tp         Vp  q         q  r         r  s         s  $u         &u  Xv         Zv  w         w  x         x  y         y  ({         *{  \|         ^|  }         }  ~         ~             ,         .  `         b             Ȅ         ʄ             0         2  d         f             ̊         Ί              4         6  h         j             А         Ґ             8                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               6!         >!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -A-    A-76-x    76-D-|    D-x-\    x--    --h    -b-    b-n-	    n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               F!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         !            B  ,          c  M            n                                    	          P	  R
          
                                   <         =  e         f                        	         
  8         9  g         h                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            9         :  s         t                        !         "  [         \             	         	  	         
  C         D  }         ~                        +         ,  e         f                                   M         N                                   5         6  o         p                        !         !  W"         X"  #         #  $          *  +          ,   -          8-  4.          l.  h/          /  0         H  J         J  VK         XK  L         L  M         M  O         O  >P         @P  xQ         zQ  R         R  S         S  &U         (U  `V         bV  W         W  X         X  Z         Z  H[         J[  \         \  ]         ]  ^         ^  0`         2`  ja         la  b         b  c         c  e         e  Rf         Tf  g         g  h         h   j         j  :k         <k  tl         vl  m         No  p         p  q         q  r         r  4t         6t  nu         pu  v         v  w         w  y         y  Vz         Xz  {         {  |         |  ~         ~  >         @  x         z                        &         (  `         b             ԇ         և             H         J                                   0         2  j         l             ޒ                              >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               F!         T!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     *
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -A-    A-;-x    ;-#J-|    #J-q}-\    q}--    --h    -ob-    ob-傳-	    傳--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               H!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             (         )  Q         R  z         {                                                           %            F  0          g  c                                              +  '           G         H  p         q               !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            3         4  g         h                                   7         8  k         l  	         	  
         
             ;         <  o         p                                   ?         @  s         t                                   C         D  w         x                                   G         H  {          |   !         !  "         "  $         $  %          &  2'         4'  f(         h(  )         )  *         *  ,         ,  6-         8-  j.         l.  /         /  0         0  2         2  :3         <3  n4         p4  5         5  6         6  
8         8  >9         @9  r:         t:  ;         ;  <         <  >         >  B?         D?  v@         x@  A         A  B         B  D         D  FE         HE  zF         |F  G         G  H          g  h          h  i          j  k          <k  >l          vl  xm         "o  Up         Vp  q         q  r         r  s         s  %u         &u  Yv         Zv  w         w  x         x  y         y  ){         *{  ]|         ^|  }         }  ~         ~             -         .  a         b             Ʉ         ʄ             1         2  e         f             ͊         Ί             5         6  i         j             ѐ         Ґ             9                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               4!         >!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     ---X     ---     -_-    _-;-x    ;-I-|    I-|-\    |--    --h    -b-    b-wx-	    wx--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               W!         X!  2         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      F         G  o	         p	  
         
           ?  g         h                                                         ,            L  $          l  D            d                                              ?   ,!          p!  ]"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            7         8  o         p                                   O         P             	         	  
         
  /         0  g         h                                   G         H                                   '         (  _         `                                   ?         @  w         x                          "          "  W#         X#  $         $  %         %  &          '  6(         8(  n)         p)  *         *  +         +  -         -  N.         P.  /         /  0         0  1         1  .3         03  f4         h4  5         5  6         6  8         8  F9         H9  ~:         :  ;         ;  <         <  &>         (>  ^?         `?  @         @  A         A  C         C  >D         @D  vE         xE  F         F  G         G  I         I  K         K  ML         NL  M         M  N         N  O         O  -Q         .Q  eR         fR  S         S  T         T  V         V  EW         FW  }X         ~X  Y         Y  Z         Z  %\         &\  ]]         ^]  ^         ^  _         _  a         a  =b         >b  uc         vc  d         d  e         e  g         g  Uh         Vh  i         i  j         j  k         k  5m         6m  mn          z  z          6{  |          ^|  >}          }  f~          ~                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               X!         ^!  2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 g
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -S-    S-;-    ;-)J-    )J-w}-d    w}-q-    q--0    --    -b-H	    b-c-P
    c-I-
    I-E׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               R!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 	          	  H
         I
  q                                 	         
  2         3  [          \  M                                              @  1          y  b                                      !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q   w  operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       \            5         6  k         l                                   C         D  y         z  	         	  
         
             Q         R                                   )         *  _         `                                   7         8  m         n                                   E         F  {         |               !         !  #         #  S$         T$  %         %  &         &  '         '  +)         ,)  a*         b*  +         +  ,         ,  .         .  9/         :/  o0         p0  1         1  2         2  4         4  G5         H5  }6         ~6  7         7  8         8  :          :  U;         V;  <         <  =         =  >         >  -@         .@  cA         dA  B         B  C         C  E         E  ;F          <F  =G          rG  sH          H  I          nn  oo          o  p         p  r         r  Ds         Fs  zt         |t  u         u  v         v  x         x  Ry         Ty  z         z  {         {  |         |  *~         ,~  `         b             ́         ΁             8         :  n         p             ڇ         ܇             F         H  |         ~                                           b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               N!         \!  1         
2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 k
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     ---X     ---     -Y-    Y-:-x    :-I-|    I-|-\    |--    --h    -b-    b-qx-	    qx--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               S!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      F         G  o	         p	  
         
           ?  g         h                                                         +            J  "          i  A            `                                  	            :   &!          k!  W"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            6         8  n         p                                   N         P             	         	  
         
  .         0  f         h                                   F         H  ~                                 &         (  ^         `                                   >         @  v         x                          "          "  V#         X#  $         $  %         %  &          '  7(         8(  o)         p)  *         *  +         +  -         -  O.         P.  /         /  0         0  1         1  /3         03  g4         h4  5         5  6         6  8         8  G9         H9  :         :  ;         ;  <         <  '>         (>  _?         `?  @         @  A         A  C         C  ?D         @D  wE         xE  F         F  G         G  I         I  K         K  LL         NL  M         M  N         N  O         O  ,Q         .Q  dR         fR  S         S  T         T  V         V  DW         FW  |X         ~X  Y         Y  Z         Z  $\         &\  \]         ^]  ^         ^  _         _  a         a  <b         >b  tc         vc  d         d  e         e  g         g  Th         Vh  i         i  j         j  k         k  4m         6m  ln                        ށ          &            N  .          v  V                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               V!         ^!  2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 g
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -S-    S-;-    ;-)J-    )J-w}-d    w}-w-    w--0    --    -b-H	    b-c-P
    c-I-
    I-E׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               Q!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 	          	  H
         I
  q                                 	         
  2         3  [          \  N                                              @  2          y  c                                      !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q   w  operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       \            4         6  j         l                                   B         D  x         z  	         	  
         
             P         R                                   (         *  ^         `                                    6         8  l         n                                   D         F  z         |               !         !  #         #  R$         T$  %         %  &         &  '         '  *)         ,)  `*         b*  +         +  ,         ,  .         .  8/         :/  n0         p0  1         1  2         2  4         4  F5         H5  |6         ~6  7         7  8         8  :          :  T;         V;  <         <  =         =  >         >  ,@         .@  bA         dA  B         B  C         C  E         E  :F         <F  qG         rG  H         H  I         nn  o         o  p          p  q          r  s          Fs  Gt          |t  }u          u  v         v  x         x  Sy         Ty  z         z  {         {  |         |  +~         ,~  a         b             ́         ΁             9         :  o         p             ۇ         ܇             G         H  }         ~                                           b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               L!         X!  1         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 k
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     ---X     ---     -Y-    Y-6-x    6-D-|    D-x-\    x-7-    7--h    -b-    b-x-	    x--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               W!         X!  2         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         !  
          B  +          c  L            m                                    	          P	  L
          
                                   <         =  e         f                                   &         '  O         P  x                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                                         0           U$  %         %  &         &  '         '  ,)         -)  b*         c*  +         +  ,         ,  .         .  :/         ;/  p0         q0  1         1  2         2  4         4  H5         I5  ~6         6  7         7  8         8   :         !:  V;         W;  <         <  =         =  >         >  .@         /@  dA         eA  B         B  C         C  E         E  <F         =F  rG         G  I         I  CJ         EJ  yK         {K  L         L  M         M  O         O  QP         SP  Q         Q  R         R  S         S  )U         +U  _V         aV  W         W  X         X  Z         Z  7[         9[  m\         o\  ]         ]  ^         ^  `         `  Ea         Ga  {b         }b  c         c  d         d  f         f  Sg         Ug  h         h  i         i  j         j  +l         l  
n         ړ             E         F  {         |                                   S         T                                   +         ,  a         b             ͤ         Τ             9         :  o         p             ۪         ܪ             G         H  }         ~                                    U         V                      µ             -                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               O!         T!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 g
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -G-    G-q;-    q;-J-    J-Y}-d    Y}--    --p    -b-    b-9-
    9-G-
    G-հ-t    հ-E׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               M!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             (         )  Q         R  z         {                                                           %            F  /          g  a                                              +  %           )         *  R         S  {          |   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         |  m                                                         T#  $         $  %         %  &         &  3(         4(  k)         l)  *         *  +         +  -         -  K.         L.  /         /  0         0  1         1  +3         ,3  c4         d4  5         5  6         6  8         8  C9         D9  {:         |:  ;         ;  <         <  #>         $>  [?         \?  @         @  A         A  C         C  ;D         <D  sE         tE  F         F  G         G  I         I  RJ         TJ  K         K  L         L  M         M  2O         4O  jP         lP  Q         Q  R         R  T         T  JU         LU  V         V  W         W  X         X  *Z         ,Z  b[         d[  \         \  ]         ]  
_         _  B`         D`  za         |a  b         b  c         c  "e         $e  Zf         \f  g         g  h         h  j         j  :k         <k  rl         tl  m         m  n         n  p         p  Rq         Tq  r         r  s         s  t         t  2v         4v  jw         lw  x         x  y         y  {         {  J|         L|  }         }  ~         ~             *         ,  b         d             ҄         Ԅ  
           B         D  z         |                        "         $  Z         \             ʏ         ̏                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               U!         f!  2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    p
         q
             R      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     &
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -S-    S-6-x    6-D-|    D-x-\    x-7-    7--h    -b-    b-x-	    x--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               S!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         !            B  ,          c  M            n                                    	          P	  K
          
                                   <         =  e         f                                   &         '  O         P  x                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         R  6          t  X            z                          !         U$  %         %  &         &  '         '  +)         -)  a*         c*  +         +  ,         ,  .         .  9/         ;/  o0         q0  1         1  2         2  4         4  G5         I5  }6         6  7         7  8         8  :         !:  U;         W;  <         <  =         =  >         >  -@         /@  cA         eA  B         B  C         C  E         E  ;F         =F  qG         G  I         I  DJ         EJ  zK         {K  L         L  M         M  O         O  RP         SP  Q         Q  R         R  S         S  *U         +U  `V         aV  W         W  X         X  Z         Z  8[         9[  n\         o\  ]         ]  ^         ^  `         `  Fa         Ga  |b         }b  c         c  d         d  f         f  Tg         Ug  h         h  i         i  j         j  ,l         l  	n         ړ             D         F  z         |                                   R         T                                   *         ,  `         b             ̤         Τ             8         :  n         p             ڪ         ܪ             F         H  |         ~                                    T         V                      µ             ,                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               K!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 g
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -G-    G-q;-    q;-J-    J-Y}-d    Y}--    --p    -b-    b-9-
    9-G-
    G-հ-t    հ-E׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               M!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             (         )  Q         R  z         {                                                           %            F  0          g  `                                              +  $           )         *  R         S  {          |   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         :  +          `  Q            w                                 T#  $         $  %         %  &         &  2(         4(  j)         l)  *         *  +         +  -         -  J.         L.  /         /  0         0  1         1  *3         ,3  b4         d4  5         5  6         6  
8         8  B9         D9  z:         |:  ;         ;  <         <  ">         $>  Z?         \?  @         @  A         A  C         C  :D         <D  rE         tE  F         F  G         G  I         I  SJ         TJ  K         K  L         L  M         M  3O         4O  kP         lP  Q         Q  R         R  T         T  KU         LU  V         V  W         W  X         X  +Z         ,Z  c[         d[  \         \  ]         ]  _         _  C`         D`  {a         |a  b         b  c         c  #e         $e  [f         \f  g         g  h         h  j         j  ;k         <k  sl         tl  m         m  n         n  p         p  Sq         Tq  r         r  s         s  t         t  3v         4v  kw         lw  x         x  y         y  {         {  K|         L|  }         }  ~         ~             +         ,  c         d             ӄ         Ԅ             C         D  {         |                        #         $  [         \             ˏ         ̏                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               U!         f!  2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    p
         q
             R      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     %
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -A-    A-:-x    :-I-|    I-|-\    |-;-    ;--h    -{b-    {b--	    --h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               I!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      F         G  o	         p	  
         
           ?  g         h                                                         %  	          >  "          W  ;          p  T                                                 !          M!  M"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            2         4  f         h                                   6         8  j         l  	         	  
         
             :         <  n         p                        
           >         @  r         t                                   B         D  v         x                                   F         H  z          |   !         !  "         "  $         $  J%         L%  ~&         &  '         '  (         (  *         *  N+         P+  ,         ,  -         -  .         .  0          0  R1         T1  2         2  3         3  4         4  "6         $6  V7         X7  8         8  9         9  :         :  &<         (<  Z=         \=  >         >  ?         ?  @         @  *B         ,B  ^C         `C  D         D  E         E  F         F  .H         0H  bI         dI  J         J  K         K  L          M  2N         4N  fO         hO  P         P  Q         Q  S         S  6T         8T  jU         lU  V         W  QX         SX  Y         Y  Z         Z  [         [  !]         #]  U^         W^  _         _  `         `  a         a  %c         'c  Yd         [d  e         e  f         f  g         g  )i         +i  ]j          <k  Bl          tl  zm          m  n          n  o          p  "q         T                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               8!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   :      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     -G-    G-)5-    )5-C-    C-w-l    w-O2-    O2--x    -b-	    b-3-
    3-A-
    A--|    -ճ-    ճ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               K!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         1  +          b  \                                                          '  	          @	  
          Y
  7         r                                              >                                 6         7  _         `                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |        "  #         #  !%         #%  [&         ]&  '         '  (         (  	*         *  C+         E+  },         ,  -         -  .         .  +0         -0  e1         g1  2         2  3         3  5         5  M6         O6  7         7  8         8  9         9  5;         7;  o<         q<  =         =  >         >  @         @  WA         YA  B         B  C         C  E         E  ?F          \  ]          ]  ^          _  _          !`  	a          Aa  )b         h  8i         :i  rj         tj  k         k  l         l   n         "n  Zo         \o  p         p  q         q  s         
s  Bt         Dt  |u         ~u  v         v  w         w  *y         ,y  dz         fz  {         {  |         |  ~         ~  L         N                                   4         6  n         p                                   V         X             ʌ         ̌             =         ?  w         y                        #         %  ]         _             ї         ӗ             E         G                                   -         /  g         i             ۢ         ݢ             O         Q             ç         ŧ             7         9  q         s                                 !  Y         [             Ͳ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               J!         T!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    p
         q
             R      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    p
         q
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     }
            0     -,-X     ,--     -A-    A-:-x    :-I-|    I-|-\    |-A-    A--h    -{b-    {b--	    --h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               I!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      F         G  o	         p	  
         
           ?  g         h                                                         %  
          >  #          W  <          p  U                                                 !          M!  L"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            3         4  g         h                                   7         8  k         l  	         	  
         
             ;         <  o         p                                   ?         @  s         t                                   C         D  w         x                                   G         H  {          |   !         !  "         "  $         $  K%         L%  &         &  '         '  (         (  *         *  O+         P+  ,         ,  -         -  .         .  0          0  S1         T1  2         2  3         3  4         4  #6         $6  W7         X7  8         8  9         9  :         :  '<         (<  [=         \=  >         >  ?         ?  @         @  +B         ,B  _C         `C  D         D  E         E  F         F  /H         0H  cI         dI  J         J  K         K  L          M  3N         4N  gO         hO  P         P  Q         Q  S         S  7T         8T  kU         lU  V         W  RX         SX  Y         Y  Z         Z  [         [  "]         #]  V^         W^  _         _  `         `  a         a  &c         'c  Zd         [d  e         e  f         f  g         g  *i         +i  ^j          Tq  Zr          r  s          s  t          t  v          4v  :w         T                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               8!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   9      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     -G-    G-)5-    )5-C-    C-w-l    w-O2-    O2--x    -b-	    b-3-
    3-A-
    A--|    -ճ-    ճ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               K!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         1  *          b  [                                                          '  	          @	  
          Y
  8         r                                              >                                 6         7  _         `                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |        "  #         #  "%         #%  \&         ]&  '         '  (         (  
*         *  D+         E+  ~,         ,  -         -  .         .  ,0         -0  f1         g1  2         2  3         3  5         5  N6         O6  7         7  8         8  9         9  6;         7;  p<         q<  =         =  >         >  @         @  XA         YA  B         B  C         C  E         E  @F          ab  Ic          c  id          d  e          e  f          f  g         h  9i         :i  sj         tj  k         k  l         l  !n         "n  [o         \o  p         p  q         q  	s         
s  Ct         Dt  }u         ~u  v         v  w         w  +y         ,y  ez         fz  {         {  |         |  ~         ~  M         N                                   5         6  o         p                                   W         X             ˌ         ̌             >         ?  x         y                        $         %  ^         _             җ         ӗ             F         G                                   .         /  h         i             ܢ         ݢ             P         Q             ħ         ŧ             8         9  r         s                                  !  Z         [             β                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               H!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    p
         q
             R      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    p
         q
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     ~
            0     -,-X     ,--     -G-    G-k;-x    k;-J-|    J-S}-\    S}--    --h    -]b-    ]b--	    --h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               O!         V!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             (         )  Q         R  z         {                       r  r                                              6  6          g  P            q                                               )         *  R         S  {          |   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            0         2  b         d                                   *         ,  \         ^  	         	  
         
             $         &  V         X                                               P         R                                              J         L  |         ~                                   D          F   v!         x!  "         "  #          /  0          0  1          1  2          3  4          D4  :5         F  G         G  &I         (I  XJ         ZJ  K         K  L         L  M         M   O         "O  RP         TP  Q         Q  R         R  S         S  U         U  LV         NV  ~W         W  X         X  Y         Y  [         [  F\         H\  x]         z]  ^         ^  _         _  a         a  @b         Bb  rc         tc  d         d  e         e  g         
g  :h         <h  li         ni  j         j  k         k  m         m  5n         6n  go         ho  p         p  q         q  r         r  /t         0t  au         bu  v         v  w         w  x         x  )z         *z  [{         \{  |         |  }         }  ~         ~  #         $  U         V                                              O         P                                              I         J  {                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               /!         6!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   :      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -M-    M-;-x    ;-J-|    J-k}-\    k}-o-    o--h    -b-    b-x-	    x--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               N!         \!  1         
2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 	          	  H
         I
  q                                 	         
  2         3  [          \  ^                                    	          @  B          y  b                                      !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            )         *  S         T  }         ~                                   %         &  O	         P	  y
         z
                                   !         "  K         L  u         v                                              G         H  q         r                                              C         D  m         n               !         !  "         $  <%         >%  f&         h&  '         '  (         (  )         )  +         +  8,         :,  b-         d-  .         .  /         /  0         0  
2         2  43         63  ^4         `4  5         5  6         6  7         7  9         9  0:         2:  Z;         \;  <         <  =         =  >         >  @         @  ,A         .A  VB         XB  C         C  D         D  E         E  F          G  (H         :H  bI         dI  J         J  K         K  L         L  
N         N  4O         6O  ^P         `P  Q         Q  R         R  S         S  U         U  0V         2V  ZW         \W  X         X  Y         Y  Z         Z  \         \  ,]         .]  V^         X^  _         _  `         `  a         a  b          c  (d         *d  Re         Te  |f         ~f  g         g  h          :i  <j          tj  vk          k  l          l  m          "n  $o         ʎ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               G!         Z!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 {
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -A-    A-k;-x    k;-J-|    J-S}-\    S}--    --h    -cb-    cb--	    --h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               K!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             (         )  Q         R  z         {                       r  q                                              6  5          g  Q            r                                               )         *  R         S  {          |   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            1         2  c         d                                   +         ,  ]         ^  	         	  
         
             %         &  W         X                                               Q         R                                              K         L  }         ~                                   E          F   w!         x!  "         "  #          l5  b6          6  7          7  8          8  9          :  ;         F  G         G  'I         (I  YJ         ZJ  K         K  L         L  M         M  !O         "O  SP         TP  Q         Q  R         R  S         S  U         U  MV         NV  W         W  X         X  Y         Y  [         [  G\         H\  y]         z]  ^         ^  _         _  a         a  Ab         Bb  sc         tc  d         d  e         e  	g         
g  ;h         <h  mi         ni  j         j  k         k  m         m  4n         6n  fo         ho  p         p  q         q  r         r  .t         0t  `u         bu  v         v  w         w  x         x  (z         *z  Z{         \{  |         |  }         }  ~         ~  "         $  T         V                                              N         P                                              H         J  z                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               1!         :!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   9      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -G-    G-;-x    ;-J-|    J-k}-\    k}-u-    u--h    -b-    b-x-	    x--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               L!         X!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 	          	  H
         I
  q                                 	         
  2         3  [          \  ]                                              @  A          y  c                                      !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            (         *  R         T  |         ~                                   $         &  N	         P	  x
         z
                                             "  J         L  t         v                                              F         H  p         r                                              B         D  l         n               !         !  "         $  =%         >%  g&         h&  '         '  (         (  )         )  +         +  9,         :,  c-         d-  .         .  /         /  0         0  2         2  53         63  _4         `4  5         5  6         6  7         7  9         9  1:         2:  [;         \;  <         <  =         =  >         >  @         @  -A         .A  WB         XB  C         C  D         D  E         E  F          G  )H         :H  cI         dI  J         J  K         K  L         L  N         N  5O         6O  _P         `P  Q         Q  R         R  S         S  U         U  1V         2V  [W         \W  X         X  Y         Y  Z         Z  \         \  -]         .]  W^         X^  _         _  `         `  a         a  b          c  )d         *d  Se         Te  }f         ~f  g         g  h          \o  ^p          p  q          q  r          
s  t          Dt  Fu         ʎ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               G!         Z!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 {
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-:-x    :-I-|    I-|-\    |-
-    
--P    -b-    b-c-	    c-a-D
    a-9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               F!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      F         G  o	         p	  
         
           ?  g         h                                                         %  	          >  "          W  ;          p  T                                                 !          M!  M"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            .         0  ^         `                                               N         P  ~	         	  
         
                        >         @  n         p                                    .         0  ^         `                                               N         P  ~                                              >!         @!  n"         %  &         &  ((         *(  X)         Z)  *         *  +         +  ,         ,  .         .  H/         J/  x0         z0  1         1  2         2  4         
4  85         :5  h6         j6  7         7  8         8  9         9  (;         *;  X<         Z<  =         =  >         >  ?         ?  A         A  HB         JB  xC         zC  D         D  E         E  G         
G  8H          8[  !\          V\  ?]          t]  ]^          ^  {_          _  `         h   j         j  0k         2k  `l         bl  m         m  n         n  o         o   q         "q  Pr         Rr  s         s  t         t  u         u  w         w  @x         Bx  py         ry  z         z  {         {   }         }  0~         2~  `         b                                              L         N  {         }             ۈ         ݈             ;                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               >!         T!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 w
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -G-    G-5-x    5-C-|    C-w-\    w--    --h    -b-    b-x-	    x--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               N!         \!  1         
2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         1  +          b  \                                                          '  	          @	  
          Y
  7         r                                              >                                 6         7  _         `                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                              @  E                     &  '         '  (         (  
*         *  6+         7+  b,         c,  -         -  .         .  /         /  1         1  >2         ?2  j3         k3  4         4  5         5  6         6  8         8  F9         G9  r:         s:  ;         ;  <         <  =         =  !?         #?  M@         O@  yA         {A  B         B  C         C  D         D  )F         +F  UG         WG  H         H  I         I  J         J  L         L  1M         3M  ]N         _N  O         O  P         P  Q         Q  S         S  9T         ;T  eU         gU  V         V  W         W  X         X  Z         Z  A[         C[  m\         o\  ]         ]  ^         ^  _         _  a         a  Ib         Kb  uc         wc  d         d  e         e  f         f  %h         'h  Qi         Si  }j         j  k         k  l          n  *o         ,o  Vp         Xp  q         q  r         r  s         s  u         u  2v         4v  ^w         `w  x         x  y         y  z         z  |         |  :}         <}  f~         h~                                              B         D  n         p             Ƈ         ȇ                         J         L  v         x             Ύ         Ў             &                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               >!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 {
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-:-x    :-{I-|    {I-|-\    |-
-    
--P    -b-    b-c-	    c-a-D
    a-9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               C!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      F         G  o	         p	  
         
           ?  g         h                                                         %  
          >  #          W  <          p  U                                                 !          M!  L"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            /         0  _         `                                               O         P  	         	  
         
                        ?         @  o         p                                    /         0  _         `                                               O         P                                                ?!         @!  o"         %  &         &  )(         *(  Y)         Z)  *         *  +         +  ,         ,  .         .  I/         J/  y0         z0  1         1  2         2  	4         
4  95         :5  i6         j6  7         7  8         8  9         9  );         *;  Y<         Z<  =         =  >         >  ?         ?  A         A  IB         JB  yC         zC  D         D  E         E  	G         
G  9H          `  a          a  b          
c  c          (d  e          Fe  /f         h  j         j  1k         2k  al         bl  m         m  n         n  o         o  !q         "q  Qr         Rr  s         s  t         t  u         u  w         w  Ax         Bx  qy         ry  z         z  {         {  }         }  1~         2~  a         b                                              M         N  |         }             ܈         ݈             <                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               >!         T!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 w
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -A-    A-5-x    5-C-|    C-w-\    w--    --h    -b-    b-x-	    x--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               L!         X!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         1  *          b  [                                                          '  	          @	  
          Y
  8         r                                              >                                 6         7  _         `                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                  @  E                                 &  '         '  (         (  	*         *  5+         7+  a,         c,  -         -  .         .  /         /  1         1  =2         ?2  i3         k3  4         4  5         5  6         6  8         8  E9         G9  q:         s:  ;         ;  <         <  =         =  "?         #?  N@         O@  zA         {A  B         B  C         C  D         D  *F         +F  VG         WG  H         H  I         I  J         J  L         L  2M         3M  ^N         _N  O         O  P         P  Q         Q  S         S  :T         ;T  fU         gU  V         V  W         W  X         X  Z         Z  B[         C[  n\         o\  ]         ]  ^         ^  _         _  a         a  Jb         Kb  vc         wc  d         d  e         e  f         f  &h         'h  Ri         Si  ~j         j  k         k  l          n  +o         ,o  Wp         Xp  q         q  r         r  s         s  u         u  3v         4v  _w         `w  x         x  y         y  z         z  |         |  ;}         <}  g~         h~                                              C         D  o         p             Ǉ         ȇ                         K         L  w         x             ώ         Ў             '                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               @!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 {
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -)-    )-_;-    _;-I-    I-G}-d    G}-!-    !--p    -b-    b-'-
    '-5-
    5--t    -3׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               A!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             (         )  Q         R  z         {                       r  r                                              6  6          g  P            q                                               )         *  R         S  {          |   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       @  K                                              %  &         &  '         '  )         
)  6*         8*  d+         f+  ,         ,  -         -  .         .  0         0  J1         L1  x2         z2  3         3  4         4  6         6  07         27  ^8         `8  9         9  :         :  ;         ;  =         =  D>         F>  r?         t?  @         @  A         A  B         B  *D         ,D  XE         ZE  F         F  G         G  H         H  J         J  >K         @K  lL         nL  M         M  N         N  O         O  %Q         &Q  SR         TR  S         S  T         T  U         U  W         W  9X         :X  gY         hY  Z         Z  [         [  \         \  ^          ^  M_         N_  {`         |`  a         a  b         b  d         d  3e         4e  af         bf  g         g  h         h  i         i  k         k  Gl         Hl  um         vm  n         n  o         o  p          q  ,r         .r  Zs         \s  t         t  u         u  v         v  x         x  @y         By  nz         pz  {         {  |         |  }         }  &         (  T         V                        ރ                      :         <  h         j             ĉ         Ɖ                       "  N         P  |         ~                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               K!         b!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    p
         q
             R      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     P
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-w;-x    w;-J-|    J-_}-\    _}-,-    ,--h    -cb-    cb-ق-	    ق--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               @!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 	          	  H
         I
  q                                 	         
  2         3  [          \  ^                                    	          @  B          y  b                                      !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            -         .  [         \                                              A         B  o	         p	  
         
                        '         (  U         V                                              ;         <  i         j                                   !         "  O         P  }         ~                        !         !  5"         6"  c#         %  &         &  '         '  %)         &)  S*         T*  +         +  ,         ,  -         -  /         /  90         :0  g1         h1  2         2  3         3  4         4  6          6  M7         N7  {8         |8  9         9  :         :  <         <  3=         4=  a>         b>  ?         ?  @         @  A         A  C         C  GD         HD  uE         vE  F         F  G         G  H         TI  J         J  K            ߁                      3            O  3          k  O           ?         A  m         o             ɐ         ː             %         '  S         U                        ݗ         ߗ             9         ;  g         i             Ý         ŝ                      !  M         O  {         }             פ         ٤             3         5  a         c                                                      >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               1!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -#-    #-_;-    _;-I-    I-G}-d    G}-!-    !--p    -b-    b-'-
    '-5-
    5--t    -3׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ?!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             (         )  Q         R  z         {                       r  q                                              6  5          g  Q            r                                               )         *  R         S  {          |   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         @  K                        	           
            @  K         %  &         &  '         '  	)         
)  7*         8*  e+         f+  ,         ,  -         -  .         .  0         0  K1         L1  y2         z2  3         3  4         4  6         6  17         27  _8         `8  9         9  :         :  ;         ;  =         =  E>         F>  s?         t?  @         @  A         A  B         B  +D         ,D  YE         ZE  F         F  G         G  H         H  J         J  ?K         @K  mL         nL  M         M  N         N  O         O  $Q         &Q  RR         TR  S         S  T         T  U         U  
W         W  8X         :X  fY         hY  Z         Z  [         [  \         \  ^          ^  L_         N_  z`         |`  a         a  b         b  d         d  2e         4e  `f         bf  g         g  h         h  i         i  k         k  Fl         Hl  tm         vm  n         n  o         o  p          q  -r         .r  [s         \s  t         t  u         u  v         v  x         x  Ay         By  oz         pz  {         {  |         |  }         }  '         (  U         V                        ߃                      ;         <  i         j             ŉ         Ɖ             !         "  O         P  }         ~                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               K!         b!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    p
         q
             R      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     Q
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-};-x    };-J-|    J-e}-\    e}-,-    ,--h    -cb-    cb-ق-	    ق--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               B!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 	          	  H
         I
  q                                 	         
  2         3  [          \  ]                                              @  A          y  c                                      !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ,         .  Z         \                                              @         B  n	         p	  
         
                        &         (  T         V                                              :         <  h         j                                             "  N         P  |         ~                        !         !  4"         6"  b#         %  &         &  '         '  $)         &)  R*         T*  +         +  ,         ,  -         -  
/         /  80         :0  f1         h1  2         2  3         3  4         4  6          6  L7         N7  z8         |8  9         9  :         :  <         <  2=         4=  `>         b>  ?         ?  @         @  A         A  C         C  FD         HD  tE         vE  F         F  G         G  H         TI  J         J  K            k                                  ۉ              ۋ           @         A  n         o             ʐ         ː             &         '  T         U                        ޗ         ߗ             :         ;  h         i             ĝ         ŝ                       !  N         O  |         }             ؤ         ٤             4         5  b         c                                                      >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               1!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     -#-    #-6-    6-D-    D-w-l    w-4-    4--x    -{b-	    {b-'-
    '-5-
    5--|    -ճ-    ճ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ?!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         !  
          B  +          c  L            m                                    	          P	  L
          
                                   <         =  e         f                                   &         '  O         P  x                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |        D%  n&          3  4          Q4  \5          5  6          6  7          8  %9         JL  uM         vM  N         N  O         O  P         P  %R         &R  QS         RS  }T         ~T  U         U  V         V  X         X  -Y         .Y  YZ         ZZ  [         [  \         \  ]         ]  	_         
_  5`         6`  aa         ba  b         b  c         c  d         d  f         f  =g         >g  ih         jh  i         i  j         j  k         k  m         m  En         Fn  qo         ro  p         p  q         q  r         r  !t         "t  Mu         Nu  yv         zv  w         w  x         x  y         y  ){         *{  U|         V|  }         }  ~         ~                        1         2  ]         ^                                              9         :  e         f                                              A         B  m         n             ē         Ɠ                        H         J  t         v             ̚         Κ             $         &  P         R  |         ~             ԡ         ֡              ,         .  X         Z                        ܨ         ި           
  4         6  `         b                                              <         >  h         j                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               @!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    p
         q
             R      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    p
         q
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            0     -,-X     ,--     -)-    )-_;-x    _;-I-|    I-G}-\    G}-{-    {--h    -Eb-    Eb-ق-	    ق--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               >!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             (         )  Q         R  z         {                                                           %            F  /          g  a                                              +  %           )         *  R         S  {          |   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            .         0  ^         `                                               N         P  ~	         	  
         
                        >         @  n         p                                    .         0  ^         `                                               N         P  ~                                              >!         @!  n"         p"  #         D%  s&         t&  '         '  (         (  *         *  3+         4+  c,         d,  -         -  .         .  /         /  #1         $1  S2         T2  3         3  4         4  5         5  7         7  C8         D8  s9         t9  :         :  ;         ;  =         =  3>         4>  c?         d?  @         @  A         A  B         B  #D         $D  SE         TE  F         F  G         G  H          H  I          "J  (K          `K  fL          L  M          M  N         ir  s         s  t         t  u         u  'w         )w  Wx         Yx  y         y  z         z  {         {  }         }  G~         I~  w         y             ׁ         ف           	  7         9  g         i             Ǉ         ɇ             '         )  W         Y                                              G         I  w         y             ה         ٔ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               (!         2!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     -#-    #-6-    6-D-    D-w-l    w-4-    4--x    -ub-	    ub-'-
    '-5-
    5--|    -ճ-    ճ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               =!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         !            B  ,          c  M            n                                    	          P	  K
          
                                   <         =  e         f                                   &         '  O         P  x                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |        D%  o&          ]9  h:          :  ;          ;  <          &=  1>          i>  t?         JL  tM         vM  N         N  O         O  P         P  $R         &R  PS         RS  |T         ~T  U         U  V         V   X         X  ,Y         .Y  XZ         ZZ  [         [  \         \  ]         ]  _         
_  4`         6`  `a         ba  b         b  c         c  d         d  f         f  <g         >g  hh         jh  i         i  j         j  k         k  m         m  Dn         Fn  po         ro  p         p  q         q  r         r   t         "t  Lu         Nu  xv         zv  w         w  x         x  y         y  ({         *{  T|         V|  }         }  ~         ~                        0         2  \         ^                                              8         :  d         f                                              @         B  l         n             œ         Ɠ                        I         J  u         v             ͚         Κ             %         &  Q         R  }         ~             ա         ֡             -         .  Y         Z                        ݨ         ި  	         
  5         6  a         b                                              =         >  i         j                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               >!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    p
         q
             R      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    p
         q
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            0     -,-X     ,--     -#-    #-_;-x    _;-I-|    I-G}-\    G}-{-    {--h    -Kb-    Kb-ق-	    ق--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               <!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             (         )  Q         R  z         {                                                           %            F  0          g  `                                              +  $           )         *  R         S  {          |   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            /         0  _         `                                               O         P  	         	  
         
                        ?         @  o         p                                    /         0  _         `                                               O         P                                                ?!         @!  o"         p"  #         D%  r&         t&  '         '  (         (  *         *  2+         4+  b,         d,  -         -  .         .  /         /  "1         $1  R2         T2  3         3  4         4  5         5  7         7  B8         D8  r9         t9  :         :  ;         ;  =         =  2>         4>  b?         d?  @         @  A         A  B         B  "D         $D  RE         TE  F         F  G         G  H          O   P          XP  ^Q          Q  R          R  S          T  U         ir  s         s  t         t  u         u  (w         )w  Xx         Yx  y         y  z         z  {         {  }         }  H~         I~  x         y             ؁         ف           	  8         9  h         i             ȇ         ɇ             (         )  X         Y                                              H         I  x         y             ؔ         ٔ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               *!         6!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-Y;-x    Y;-I-|    I-A}-\    A}-?6-    ?6--h    -b-    b-m-	    m--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               @!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             (         )  Q         R  z         {                       r  r                                              6  6          g  P            q                                               )         *  R         S  {          |   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            (         *  R         T  |         ~                                   $         &  N	         P	  x
         z
                                             "  J         L  t         v                                              F         H  p         r                                              B         D  l         n               !         !  "          ;  <          <  =          (>  "?          Z?  T@          @  A         K  L         L  N         N  1O         2O  [P         \P  Q         Q  R         R  S         S  U         U  -V         .V  WW         XW  X         X  Y         Y  Z         Z  [          \  )]         *]  S^         T^  }_         ~_  `         `  a         a  b         b  %d         &d  Oe         Pe  yf         zf  g         g  h         h  i         i  !k         "k  Kl         Ll  um         vm  n         n  o           6         8  `         b                        ޚ                    
  2         4  \         ^                        ڡ         ܡ             .         0  X         Z                        ֨         ب              *         ,  T         V  ~                      ү         ԯ             &         (  P         R  z         |             ζ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               Y!         Z!  2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     +
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-q;-x    q;-J-|    J-Y}-\    Y}-c$-    c$--P    -b-    b-c-	    c-7-D
    7--׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               K!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 	          	  H
         I
  q                                 	         
  2         3  [          \  ^                                    	          @  B          y  b                                      !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            1         2  c         d                                   +         ,  ]         ^  	         	  
         
             %         &  W         X                                               Q         R                                              K         L  }         ~                                   E          F   w!         x!  "         "  #         #  %         %  ?&         @&  q'         r'  (         (  )         )  +         +  9,         :,  k-         l-  .         .  /         /  1         1  32         42  e3         f3  4         4  5         5  6         6  -8         .8  _9         `9  :         :  ;         ;  <         <  '>         (>  Y?         Z?  @         @  A         A  B         B  !D         "D  SE         TE  F         F  G         G  H         H  J         ir  s         s  t         t  u         u  /w         1w  ax         cx  y         y  z         z  {         {  )}         +}  [~         ]~                                   #         %  U         W                                              O         Q                                              I          ޞ            !  ,          d  o                                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               O!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 k
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -] -X     ] -հ-     հ-%ܰ-    %ܰ-c4-p    c4-!@-T    !@-h-4    h-?6-    ?6--8    -b-    b-m-	    m--8
    --׳-
                   =         b      J      7   U   7   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               B!         T!  P1                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 64 })  outputElements  5184    input0Elements  31104   input1Elements  24576   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     4                           #         b      J      7   M     operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       @             (         )  Q         R  z         {            (	  '
          Y
  X                                                          .            O  9         4  \         ]                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     4                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  1728    input0Elements  1728    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       x            )         *  S         T  }         ~                                   %         &  O	         P	  y
         z
                                   !         "  K         L  u         v                                              G         H  q         r                                              C         D  m         n               !         !  "          A  B          B  C          "D  E          TE  NF         K  L         L  N         N  0O         2O  ZP         \P  Q         Q  R         R  S         S  U         U  ,V         .V  VW         XW  X         X  Y         Y  Z         Z  [          \  (]         *]  R^         T^  |_         ~_  `         `  a         a  b         b  $d         &d  Ne         Pe  xf         zf  g         g  h         h  i         i   k         "k  Jl         Ll  tm         vm  n         n  o           7         8  a         b                        ߚ           	         
  3         4  ]         ^                        ۡ         ܡ             /         0  Y         Z                        ר         ب             +         ,  U         V                        ӯ         ԯ             '         (  Q         R  {         |             ϶                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               Y!         Z!  2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     *
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-q;-x    q;-J-|    J-Y}-\    Y}-*-    *--P    -b-    b-c-	    c-7-D
    7--׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               H!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 	          	  H
         I
  q                                 	         
  2         3  [          \  ]                                              @  A          y  c                                      !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            0         2  b         d                                   *         ,  \         ^  	         	  
         
             $         &  V         X                                               P         R                                              J         L  |         ~                                   D          F   v!         x!  "         "  #         #  %         %  >&         @&  p'         r'  (         (  )         )  +         +  8,         :,  j-         l-  .         .  /         /   1         1  22         42  d3         f3  4         4  5         5  6         6  ,8         .8  ^9         `9  :         :  ;         ;  <         <  &>         (>  X?         Z?  @         @  A         A  B         B   D         "D  RE         TE  F         F  G         G  H         H  J         ir  s         s  t         t  u         u  0w         1w  bx         cx  y         y  z         z  {         {  *}         +}  \~         ]~                                   $         %  V         W                                              P         Q                                              J          -  8          p  {                                  9  D                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               O!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 k
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -{,-X     {,-1-     1--    -6-x    6-D-|    D-x-\    x-1-    1--h    -ob-    ob-m-	    m--h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     X                       H!  1          1  t=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     >  #          ]  B          |  a                                    	          L	  P
          
                                   B         C  m         n                                   @         A  k         l                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            4         6  j         l                                   B         D  x         z  	         	  
         
             P         R                                   (         *  ^         `                                    6         8  l         n                                   D         F  z         |               !         !  #         #  R$          $  |%          %  &          &  '          (  (          ()  *          I  UJ         VJ  K         K  L         L  M         M  -O         .O  cP         dP  Q         Q  R         R  T         T  ;U         <U  qV         rV  W         W  X         X  Z         Z  I[         J[  \         \  ]         ]  ^         ^  !`         "`  Wa         Xa  b         b  c         c  d         d  /f         0f  eg         fg  h         h  i         i  k         k  =l         >l  sm         tm  n         n  o         o  q         q  Kr         Lr  s         s  t         t  u         u  #w         $w  Yx         Zx  y         y  z         z  {         {  1}         2}  g~         h~             Ӏ         Ԁ  	         
  ?         @  u         v                                   M         N                                   %         &  [         \             Ǒ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               G!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     +
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,-[-     [--    -;;-x    ;;-I-|    I-#}-\    #}--    --h    -Wb-    Wb-ǂ-	    ǂ--h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     i            !          V!  1          2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             *         +  U         V                                                                                   >  #          ]  [                                                                      8         9  c          d   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            6         8  n         p                                   N         P             	         	  
         
  .         0  f         h                                   F         H  ~                                 &         (  ^         `                                   >         @  v         x                          "          "  V#         X#  $         $  %         %  &          '  7(         8(  o)         p)  *         *  +         +  -         -  O.         P.  /         /  0         0  1         1  /3         03  g4         h4  5         5  6         6  8         8  G9         H9  :         :  ;         ;  <         <  '>         (>  _?         `?  @         @  A         A  C         C  ?D         @D  wE         xE  F         F  G         G  I          vL  M          M  N          N  P          ?P  HQ          Q  R         o   q         q  8r         :r  ps         rs  t         t  u         u  w         w  Px         Rx  y         y  z         z  {         {  0}         2}  h~         j~             ؀         ڀ             H         J                                   (         *  `         b             Ћ         ҋ           
  @         B  x         z                                  "  X                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               9!         :!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                          
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -{,-X     {,-1-     1--    -6-x    6-D-|    D-x-\    x-1-    1--h    -ob-    ob-m-	    m--h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     Y                       H!  1          1  u=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     >  $          ]  C          |  b                                    	          L	  O
          
                                   B         C  m         n                                   @         A  k         l                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            5         6  k         l                                   C         D  y         z  	         	  
         
             Q         R                                   )         *  _         `                                   7         8  m         n                                   E         F  {         |               !         !  #         #  S$          N*  :+          t+  `,          ,  -          -  .          .  /          I  TJ         VJ  K         K  L         L  M         M  ,O         .O  bP         dP  Q         Q  R         R  T         T  :U         <U  pV         rV  W         W  X         X  Z         Z  H[         J[  ~\         \  ]         ]  ^         ^   `         "`  Va         Xa  b         b  c         c  d         d  .f         0f  dg         fg  h         h  i         i  k         k  <l         >l  rm         tm  n         n  o         o  q         q  Jr         Lr  s         s  t         t  u         u  "w         $w  Xx         Zx  y         y  z         z  {         {  0}         2}  f~         h~             Ҁ         Ԁ           
  >         @  t         v                                   L         N                                   $         &  Z         \             Ƒ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               G!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     *
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,-U-     U--    -;;-x    ;;-I-|    I-#}-\    #}--    --h    -Wb-    Wb-ǂ-	    ǂ--h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     h            !          V!  1          2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             *         +  U         V                                                                                   >  $          ]  Z                                                                      8         9  c          d   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            7         8  o         p                                   O         P             	         	  
         
  /         0  g         h                                   G         H                                   '         (  _         `                                   ?         @  w         x                          "          "  W#         X#  $         $  %         %  &          '  6(         8(  n)         p)  *         *  +         +  -         -  N.         P.  /         /  0         0  1         1  .3         03  f4         h4  5         5  6         6  8         8  F9         H9  ~:         :  ;         ;  <         <  &>         (>  ^?         `?  @         @  A         A  C         C  >D         @D  vE         xE  F         F  G         G  I          R  S          T  U          KU  TV          V  W          W  X         o  q         q  9r         :r  qs         rs  t         t  u         u  w         w  Qx         Rx  y         y  z         z  {         {  1}         2}  i~         j~             ـ         ڀ             I         J                                   )         *  a         b             ы         ҋ  	         
  A         B  y         z                        !         "  Y                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               9!         :!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                          
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,-=-     =--    -:-x    :-{I-|    {I-|-\    |--    --h    -ub-    ub-Yx-	    Yx--h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     Z             !          L!  1          1  |=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      @         A  k	         l	  
         
           I  s         t                                                          ;            V  2          q  M            h                                              4   (!          c!  W"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            2         4  f         h                                   6         8  j         l  	         	  
         
             :         <  n         p                        
           >         @  r         t                                   B         D  v         x                                   F         H  z          |   !         !  "         "  $         $  J%         L%  ~&         &  '         '  (         (  *         *  N+         P+  ,         ,  -         -  .         .  0          0  R1         T1  2         2  3         3  4         4  "6         $6  V7         X7  8         8  9         9  :         :  &<         (<  Z=         \=  >         >  ?         ?  @         @  *B         ,B  ^C         `C  D         D  E         E  F         F  .H         ^H  I         I  J         J  K         K  -M         .M  aN         bN  O         O  P         P  Q         Q  1S         2S  eT         fT  U         U  V         V  X         X  5Y         6Y  iZ         jZ  [         [  \         \  ^         ^  9_         :_  m`         n`  a         a  b         b  	d         
d  =e         >e  qf         rf  g         g  h         h  j         j  Ak         Bk  ul          y  z          z  {          {  |          2}  )~          h~  _                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               H!         J!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 e
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            4     -,-`     ,-a-     a--    -;-    ;-J-    J-k}-d    k}-7-    7--0    --    -b-H	    b-c-P
    c-%-
    %--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     i            !          X!  1          2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 	         	  F
         G
  q                                            D         E  o          p  j                                  !            \  V            |                                      !          "  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q   w  operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       \            9         :  s         t                        !         "  [         \             	         	  	         
  C         D  }         ~                        +         ,  e         f                                   M         N                                   5         6  o         p                        !         !  W"         X"  #         #  $         $  &         &  ?'         @'  y(         z(  )         )  *         *  ',         (,  a-         b-  .         .  /         /  1         1  I2         J2  3         3  4         4  5         5  17         27  k8         l8  9         9  :         :  <         <  S=         T=  >         >  ?         ?  A         A  ;B         <B  uC         vC  D         D  E         E  #G         $G  \H         vl  m         m  n         n  "p         $p  \q         ^q  r         r  s         s  
u         u  Dv         Fv  ~w          w  x          x  y          y  z          .{  1|          h|  k}         }  ~         ~             N         P                      ă             6         8  p         r                                    X         Z             ̎                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               N!         X!  1         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 i
                     
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,-7-     7--    -:-x    :-{I-|    {I-|-\    |--    --h    -{b-    {b-Yx-	    Yx--h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     Y                       H!  1          1  u=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      @         A  k	         l	  
         
           I  s         t                                                          ;            V  3          q  N            i                                              4   '!          c!  V"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            3         4  g         h                                   7         8  k         l  	         	  
         
             ;         <  o         p                                   ?         @  s         t                                   C         D  w         x                                   G         H  {          |   !         !  "         "  $         $  K%         L%  &         &  '         '  (         (  *         *  O+         P+  ,         ,  -         -  .         .  0          0  S1         T1  2         2  3         3  4         4  #6         $6  W7         X7  8         8  9         9  :         :  '<         (<  [=         \=  >         >  ?         ?  @         @  +B         ,B  _C         `C  D         D  E         E  F         F  /H         ^H  I         I  J         J  K         K  ,M         .M  `N         bN  O         O  P         P  Q         Q  0S         2S  dT         fT  U         U  V         V   X         X  4Y         6Y  hZ         jZ  [         [  \         \  ^         ^  8_         :_  l`         n`  a         a  b         b  d         
d  <e         >e  pf         rf  g         g  h         h  j         j  @k         Bk  tl                      Ԁ  ˁ          
            @  7          v  m                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               J!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 e
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            4     -,-`     ,-[-     [--    -;-    ;-J-    J-k}-d    k}-7-    7--0    --    -b-H	    b-c-P
    c-%-
    %--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     h            !          V!  1          2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 	         	  F
         G
  q                                            D         E  o          p  i                                  !            \  U            }                                      !          "  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q   w  operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       \            8         :  r         t                                  "  Z         \             	         	           
  B         D  |         ~                        *         ,  d         f                                   L         N                                   4         6  n         p                        !         !  V"         X"  #         #  $         $  &         &  >'         @'  x(         z(  )         )  *         *  &,         (,  `-         b-  .         .  /         /  1         1  H2         J2  3         3  4         4  5         5  07         27  j8         l8  9         9  :         :  <         <  R=         T=  >         >  ?         ?   A         A  :B         <B  tC         vC  D         D  E         E  "G         $G  ]H         vl  m         m  n         n  #p         $p  ]q         ^q  r         r  s         s  u         u  Ev         Fv  w         w  x         x  y         y  -{         .{  g|         h|  }          }  ~          ~                        P  S                     ă             7         8  q         r                                    Y         Z             ͎                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               L!         T!  1         1  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 i
                     
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,-y-     y--    -%6-x    %6-D-|    D-x-\    x--    -}-h    }-b-    b-_x-	    _x--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               7!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     >  #          ]  B          |  a                                    	          L	  P
          
                                   B         C  m         n                                   @         A  k         l                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                              P  H            x         $  %         %  '         '  8(         :(  j)         l)  *         *  +         +   -         -  2.         4.  d/         f/  0         0  1         1  2         2  ,4         .4  ^5         `5  6         6  7         7  8         8  &:         (:  X;         Z;  <         <  =         =  >         >   @         "@  RA         TA  B         B  C         C  D         D  F         F  LG         NG  ~H         H  I         I  K         K  BL         DL  tM         vM  N         N  O         O  
Q         Q  <R         >R  nS         pS  T         T  U         U  W         W  7X         8X  iY         jY  Z         Z  [         [  \          ]  1^         2^  c_         d_  `         `  a         a  b         b  +d         ,d  ]e         ^e  f         f  g         g  h         h  %j         &j  Wk         Xk  l         l  m         m  n         n  p          p  Qq         Rq  r         r  s         s  t         t  v         v  Kw         Lw  }x         ~x  y         y  z         z  |         |  E}         F}  w~         x~             ۀ         ܀             ?         @  q         r             Ն         ֆ             9         :  k         l             ό         Ќ             3         d                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     X                       <!  1          1  b=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 e
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -A-    A-G;-    G;-I-    I-/}-d    /}--    --p    -ob-    ob-'-
    '-5-
    5--t    --׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               L!         `!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             *         +  U         V                                                                                   >  #          ]  [                                                                      8         9  c          d   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         H  ;          r  e                                             T#  $         $  %         %  '         '  C(         D(  )         )  *         *  +         +  3-         4-  o.         p.  /         /  0         0  #2         $2  _3         `3  4         4  5         5  7         7  O8         P8  9         9  :         :  <         <  ?=         @=  {>         |>  ?         ?  @         @  /B         0B  kC         lC  D         D  E         E  G          G  [H         \H  I         J  K         K  8M         :M  tN         vN  O         O  P         P  (R         *R  dS         fS  T         T  U         U  W         W  TX         VX  Y         Y  Z         Z  \         
\  D]         F]  ^         ^  _         _  `         `  4b         6b  pc         rc  d         d  e         e  $g         &g  `h         bh  i         i  j         j  l         l  Pm         Rm  n         n  o         o  q         q  @r         Br  |s         ~s  t         t  u         u  0w         2w  lx         nx  y         y  z         z   |         "|  \}         ^}  ~         ~                        L         N             Ą         Ƅ              <         >  x         z                        ,         .  h         j                                   X         Z                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     i            !          j!  1          2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    r
         s
             X      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                      
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-s-     s--    -%6-x    %6-D-|    D-x-\    x--    --h    -!b-    !b-_x-	    _x--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               5!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     >  $          ]  C          |  b                                    	          L	  O
          
                                   B         C  m         n                                   @         A  k         l                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                    !          @!  8"          p"  h#         $  %         %  '         '  9(         :(  k)         l)  *         *  +         +  -         -  3.         4.  e/         f/  0         0  1         1  2         2  -4         .4  _5         `5  6         6  7         7  8         8  ':         (:  Y;         Z;  <         <  =         =  >         >  !@         "@  SA         TA  B         B  C         C  D         D  F         F  MG         NG  H         H  I         I  K         K  CL         DL  uM         vM  N         N  O         O  Q         Q  =R         >R  oS         pS  T         T  U         U  W         W  6X         8X  hY         jY  Z         Z  [         [  \          ]  0^         2^  b_         d_  `         `  a         a  b         b  *d         ,d  \e         ^e  f         f  g         g  h         h  $j         &j  Vk         Xk  l         l  m         m  n         n  p          p  Pq         Rq  r         r  s         s  t         t  v         v  Jw         Lw  |x         ~x  y         y  z         z  |         |  D}         F}  v~         x~             ڀ         ܀             >         @  p         r             Ԇ         ֆ             8         :  j         l             Ό         Ќ              2         d                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     Y                       @!  1          1  i=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 e
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -;-    ;-G;-    G;-I-    I-/}-d    /}--    --p    -ob-    ob-'-
    '-5-
    5--t    --׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               H!         \!  1         
2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             *         +  U         V                                                                                   >  $          ]  Z                                                                      8         9  c          d   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                     D  7          n  a              !         T#  $         $  %         %  '         '  B(         D(  ~)         )  *         *  +         +  2-         4-  n.         p.  /         /  0         0  "2         $2  ^3         `3  4         4  5         5  7         7  N8         P8  9         9  :         :  <         <  >=         @=  z>         |>  ?         ?  @         @  .B         0B  jC         lC  D         D  E         E  G          G  ZH         \H  I          I  J         J  K         K  9M         :M  uN         vN  O         O  P         P  )R         *R  eS         fS  T         T  U         U  W         W  UX         VX  Y         Y  Z         Z  	\         
\  E]         F]  ^         ^  _         _  `         `  5b         6b  qc         rc  d         d  e         e  %g         &g  ah         bh  i         i  j         j  l         l  Qm         Rm  n         n  o         o  q         q  Ar         Br  }s         ~s  t         t  u         u  1w         2w  mx         nx  y         y  z         z  !|         "|  ]}         ^}  ~         ~                        M         N             ń         Ƅ             =         >  y         z                        -         .  i         j                                   Y         Z                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     h            !          j!  1          2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    r
         s
             X      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                      
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-I-     I--    -:-x    :-I-|    I-|-\    |-1-    1--h    --b-    -b--	    --h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               !         0!  1         1  ~=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      @         A  k	         l	  
         
           I  s         t                                                          ;            V  2          q  M            h                                              4   (!          c!  W"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            .         0  ^         `                                               N         P  ~	         	  
         
                        >         @  n         p                                    .         0  ^         `                                               N         P  ~                                              >!         @!  n"         p"  #         #  $         $  %          &  .'         0'  ^(         `(  )         )  *         *  +         +  -          -  N.         P.  ~/         /  0         0  1         1  3         3  >4         @4  n5         p5  6         6  7         7  8          9  .:         0:  ^;         `;  <         <  =         =  >         >  @          @  NA         PA  ~B         B  C         C  D         D  F         F  >G         @G  nH         pH  I         I  J          J  K          K  L          $M  N          NN  <O          xO  fP         d                      Ĕ             "         $  R         T                                              B         D  r         t             ҡ         ԡ             2         4  b         d             §         ħ             "         $  R         T                                              B         D  r                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     U                       H!  1          1  q=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   *      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -;-    ;-55-x    55-C-|    C-w-\    w-	-    	--h    -b-    b-!n-	    !n--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               H!         X!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         /  -          ^  \                                                          !  	          <	  "
          W
  =         r                                              H                      %         &  P         Q  {         |                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            *         ,  V         X                                              2         4  ^	         `	  
         
                                   :         <  f         h                                              B         D  n         p                                               J         L  v         x                          !         !  &#          d#  `$          $  %          %  &          &  '          (  )         HI  rJ         tJ  K         K  L         L  M         M  "O         $O  NP         PP  zQ         |Q  R         R  S         S  T          U  *V         ,V  VW         XW  X         X  Y         Y  Z         Z  \         \  2]         4]  ^^         `^  _         _  `         `  a         a  c         c  :d         <d  fe         he  f         f  g         g  h         h  j         j  Bk         Dk  nl         pl  m         m  n         n  o         o  q          q  Jr         Lr  vs         xs  t         t  u         u  v         v  &x         (x  Ry         Ty  ~z         z  {         {  |         |  ~         ~  .         0  Z         \                        ރ           
           6         8  b         d                                              >         @  j         l                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               H!         X!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     O
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-C-     C--    -:-x    :-I-|    I-|-\    |-1-    1--h    -'b-    'b--	    --h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               !         .!  1         1  {=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      @         A  k	         l	  
         
           I  s         t                                                          ;            V  3          q  N            i                                              4   '!          c!  V"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            /         0  _         `                                               O         P  	         	  
         
                        ?         @  o         p                                    /         0  _         `                                               O         P                                                ?!         @!  o"         p"  #         #  $         $  %          &  /'         0'  _(         `(  )         )  *         *  +         +  -          -  O.         P.  /         /  0         0  1         1  3         3  ?4         @4  o5         p5  6         6  7         7  8          9  /:         0:  _;         `;  <         <  =         =  >         >  @          @  OA         PA  B         B  C         C  D         D  F         F  ?G         @G  oH         pH  I         I  J          P  Q          Q  R          R  S           T  U          JU  8V         d             Ô         Ĕ             #         $  S         T                                              C         D  s         t             ӡ         ԡ             3         4  c         d             ç         ħ             #         $  S         T                                              C         D  s                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     T                       D!  1          1  j=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   )      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-55-x    55-C-|    C-w-\    w-	-    	--h    -b-    b-!n-	    !n--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               D!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         /  ,          ^  [                                                          !  	          <	  #
          W
  >         r                                              H                      %         &  P         Q  {         |                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            +         ,  W         X                                              3         4  _	         `	  
         
                                   ;         <  g         h                                              C         D  o         p                                               K         L  w         x                          !         !  '#          J)  F*          x*  t+          +  ,          ,  -          .  .         HI  sJ         tJ  K         K  L         L  M         M  #O         $O  OP         PP  {Q         |Q  R         R  S         S  T          U  +V         ,V  WW         XW  X         X  Y         Y  Z         Z  \         \  3]         4]  _^         `^  _         _  `         `  a         a  c         c  ;d         <d  ge         he  f         f  g         g  h         h  j         j  Ck         Dk  ol         pl  m         m  n         n  o         o  q          q  Kr         Lr  ws         xs  t         t  u         u  v         v  'x         (x  Sy         Ty  z         z  {         {  |         |  ~         ~  /         0  [         \                        ߃                      7         8  c         d                                              ?         @  k         l                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               D!         T!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     N
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-M;-x    M;-I-|    I-5}-\    5}--    --h    -ub-    ub--	    --h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ?!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             *         +  U         V                                  r  f                                              .  "          ]  B          |  a                                                          8         9  c          d   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ,         .  Z         \                                              @         B  n	         p	  
         
                        &         (  T         V                                              :         <  h         j                                             "  N         P  |         ~                        !         !  4"         6"  b#         d#  $         $  %         %  &         &  (         (  H)         J)  v*         x*  +         +  ,         ,   .         .  ./         0/  \0         ^0  1         1  2         2  3         3  5         5  B6         D6  p7         r7  8         8  9         9  :         :  (<         *<  V=         X=  >         >  ?         ?  @         @  B         B  <C         >C  jD         lD  E         E  F          R  R          3S  T          TT  9U          uU  ZV          V  {W         l  m         m  o          o  Mp         Np  {q         |q  r         r  s         s  u         u  3v         4v  aw         bw  x         x  y         y  z         z  |         |  G}         H}  u~         v~             р         Ҁ              -         .  [         \                                              A         B  o         p             ˍ         ̍             '                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               3!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   *      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -;-    ;-;-x    ;-)J-|    )J-w}-\    w}-m-    m--h    -b-    b-x-	    x--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               H!         X!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 	         	  F
         G
  q                                            D         E  o          p  j                                  !            \  V            |                                      !          "  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            -         .  [         \                                              A         B  o	         p	  
         
                        '         (  U         V                                              ;         <  i         j                                   !         "  O         P  }         ~                        !         !  5"         6"  c#         d#  $         $  %         %  &         &  (         (  H)         J)  v*         x*  +         +  ,         ,   .         .  ./         0/  \0         ^0  1         1  2         2  3         3  5         5  B6         D6  p7         r7  8         8  9         9  :         :  (<         *<  V=         X=  >         >  ?         ?  @         @  B         B  <C         >C  jD         lD  E         E  F         F  G         G  "I         $I  PJ         RJ  ~K         K  L         L  M         M  O         
O  6P         8P  dQ         fQ  R         R  S         S  T         T  V         V  JW         LW  xX         zX  Y         Y  Z         Z  \         \  0]         2]  ^^         `^  _         _  `         `  a         a  c         c  Dd         Fd  re         te  f         f  g         g  h          0y  ,z          nz  j{          {  |          |  }          (~  $         ʐ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               K!         ^!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 y
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-M;-x    M;-I-|    I-5}-\    5}--    --h    -ub-    ub--	    --h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               C!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             *         +  U         V                                  r  e                                              .  !          ]  C          |  b                                                          8         9  c          d   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            -         .  [         \                                              A         B  o	         p	  
         
                        '         (  U         V                                              ;         <  i         j                                   !         "  O         P  }         ~                        !         !  5"         6"  c#         d#  $         $  %         %  &         &  (         (  I)         J)  w*         x*  +         +  ,         ,  .         .  //         0/  ]0         ^0  1         1  2         2  3         3  5         5  C6         D6  q7         r7  8         8  9         9  :         :  )<         *<  W=         X=  >         >  ?         ?  @         @  B         B  =C         >C  kD         lD  E         E  F          W  X          X  Y          Y  Z          [  [          ;\   ]         l  m         m  o          o  Lp         Np  zq         |q  r         r  s         s  u         u  2v         4v  `w         bw  x         x  y         y  z         z  |         |  F}         H}  t~         v~             Ѐ         Ҁ              ,         .  Z         \                                              @         B  n         p             ʍ         ̍             &                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               3!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   )      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-;-x    ;-)J-|    )J-w}-\    w}-m-    m--h    -b-    b-x-	    x--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               D!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 	         	  F
         G
  q                                            D         E  o          p  i                                  !            \  U            }                                      !          "  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ,         .  Z         \                                              @         B  n	         p	  
         
                        &         (  T         V                                              :         <  h         j                                             "  N         P  |         ~                        !         !  4"         6"  b#         d#  $         $  %         %  &         &  (         (  I)         J)  w*         x*  +         +  ,         ,  .         .  //         0/  ]0         ^0  1         1  2         2  3         3  5         5  C6         D6  q7         r7  8         8  9         9  :         :  )<         *<  W=         X=  >         >  ?         ?  @         @  B         B  =C         >C  kD         lD  E         E  F         F  G         G  #I         $I  QJ         RJ  K         K  L         L  M         M  	O         
O  7P         8P  eQ         fQ  R         R  S         S  T         T  V         V  KW         LW  yX         zX  Y         Y  Z         Z  \         \  1]         2]  _^         `^  _         _  `         `  a         a  c         c  Ed         Fd  se         te  f         f  g         g  h          f  b                        ނ                       ^  Z         ʐ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               I!         ^!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 y
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -/-    /-:-    :-I-    I-|-d    |-q
-    q
--0    --    -b-H	    b-c-P
    c-I-
    I-9׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               @!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      @         A  k	         l	  
         
           I  s         t                                                          ;            V  2          q  M            h                                              4   (!          c!  W"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q   w  operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       \            *         ,  V         X                                              2         4  ^	         `	  
         
                                   :         <  f         h                                              B         D  n         p                                               J         L  v         x                          !         !  &#         (#  R$         T$  ~%         %  &         &  '         '  )         )  .*         0*  Z+         \+  ,         ,  -         -  .         .  
0         0  61         81  b2         d2  3         3  4         4  5         5  7         7  >8         @8  j9         l9  :         :  ;         ;  <         <  >         >  F?         H?  r@         t@  A         A  B         B  C         h  (j         *j  Tk         Vk  l         l  m         m  n         n  p         p  0q         2q  \r         ^r  s         s  t         t  u         u  w         w  8x         :x  dy         fy  z         z  {         {  |         |  ~         ~  @         B  l          n  b                      Ă                                   E  o         q             ǉ         ɉ                                b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               H!         l!  1         "2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 u
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -A-    A-A5-x    A5-C-|    C-)w-\    )w-i0-    i0--h    -b-    b-x-	    x--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               J!         X!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         /  -          ^  \                                                          !  	          <	  "
          W
  =         r                                              H                      %         &  P         Q  {         |                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         <  J                                              H  V         %  	'         
'  9(         :(  i)         j)  *         *  +         +  ,         ,  ).         *.  Y/         Z/  0         0  1         1  2         2  4         4  I5         J5  y6         z6  7         7  8         8  	:         
:  9;         :;  i<         j<  =         =  >         >  ?         ?  )A         *A  YB         ZB  C         C  D         D  E         E  G         G  IH         JH  yI         h  i         i  *k         ,k  Zl         \l  m         m  n         n  o         o  q         q  Jr         Lr  zs         |s  t         t  u         u  
w         w  :x         <x  jy         ly  z         z  {         {  |         |  *~         ,~  Z         \                                              J         L  z         |             ڈ         ܈  
           :         <  j         P  ~                      ސ                      >         @  n         p             Ζ         Ж              .         0  ^         `                                               N         P  ~                      ޣ                      >         @  n         p             Ω         Щ              .         0  ^         `                                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               F!         T!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 y
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -o -`     o -ְ-     ְ-7ܰ-    7ܰ-4-x    4-?-\    ?-)h-<    )h-;
-    ;
-e-    e--    -b- 	    b-c-(
    c-I-
    I-9׳-
                   =         b      J      7   U   7   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               C!         R!  P1                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 64 })  outputElements  5184    input0Elements  31104   input1Elements  24576   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     4                           #         b      J      7   M     operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       @                                 <         =  g	                      )         *  T         U              ]            x                                                          J  =          y  l                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     4                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  1728    input0Elements  1728    invocations     1                           3         b      J      7   Q   w  operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       \            +         ,  W         X                                              3         4  _	         `	  
         
                                   ;         <  g         h                                              C         D  o         p                                               K         L  w         x                          !         !  '#         (#  S$         T$  %         %  &         &  '         '  )         )  /*         0*  [+         \+  ,         ,  -         -  .         .  0         0  71         81  c2         d2  3         3  4         4  5         5  7         7  ?8         @8  k9         l9  :         :  ;         ;  <         <  >         >  G?         H?  s@         t@  A         A  B         B  C         h  )j         *j  Uk         Vk  l         l  m         m  n         n  p         p  1q         2q  ]r         ^r  s         s  t         t  u         u  w         w  9x         :x  ey         fy  z         z  {         {  |         |  ~         ~  A         B  m         n             Â         Ă                        D          E  :          q  f                      ɉ                                b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     4                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               F!         h!  1         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 u
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -;-    ;-A5-x    A5-C-|    C-)w-\    )w-i0-    i0--h    -b-    b-x-	    x--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               F!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         /  ,          ^  [                                                          !  	          <	  #
          W
  >         r                                              H                      %         &  P         Q  {         |                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                         !          "  #          T#  b$          $  %         %  '         
'  8(         :(  h)         j)  *         *  +         +  ,         ,  (.         *.  X/         Z/  0         0  1         1  2         2  4         4  H5         J5  x6         z6  7         7  8         8  :         
:  8;         :;  h<         j<  =         =  >         >  ?         ?  (A         *A  XB         ZB  C         C  D         D  E         E  G         G  HH         JH  xI         h  i         i  +k         ,k  [l         \l  m         m  n         n  o         o  q         q  Kr         Lr  {s         |s  t         t  u         u  w         w  ;x         <x  ky         ly  z         z  {         {  |         |  +~         ,~  [         \                                              K         L  {         |             ۈ         ܈             ;         <  k         P                        ߐ                      ?         @  o         p             ϖ         Ж              /         0  _         `                                               O         P                        ߣ                      ?         @  o         p             ϩ         Щ              /         0  _         `                                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               B!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 y
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -A-    A-S;-x    S;-I-|    I-;}-\    ;}-!9-    !9--h    -b-    b-	n-	    	n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               D!         X!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             *         +  U         V                                  r  f                                              .  "          ]  B          |  a                                                          8         9  c          d   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            :         <  v         x                        *         ,  f         h             	         	             V         X                        
           F         H                                   6         8  r         t                        &         (  b         d                                     R!         T!  "         "  #         #  %         %  C&         D&  '         '  (         (  )         )  3+         4+  o,         p,  -         -  .         .  #0         $0  _1         `1  2         2  3         3  5         5  O6         P6  7         7  8         8  :         :  ?;         @;  {<         |<  =         =  >         >  /@         0@  kA         lA  B         B  C         C  E          E  [F         \F  G          G  H          .I  7J          qJ  zK          K  L          L   N         m  o         o  Op           ֕         ؕ             N         P             ƚ         Ț             >         @  z         |                        .         0  j         l                                    Z         \             Ҫ         Ԫ             J         L             ¯         į              :         <  v         x                        *         ,  f         h                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               I!         b!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     +
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-;-x    ;-/J-|    /J-}}-\    }}-I-    I--h    -ub-    ub-߂-	    ߂--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               B!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 	         	  F
         G
  q                                            D         E  o          p  j                                  !            \  V            |                                      !          "  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            1         2  c         d                                   +         ,  ]         ^  	         	  
         
             %         &  W         X                                               Q         R                                              K         L  }         ~                                   E          F   w!         x!  "         "  #          %  %          2&  "'          \'  L(          (  v)          )  *         G  H         H  I         I  -K         .K  _L         `L  M         M  N         N  O         O  'Q         (Q  YR         ZR  S         S  T         T  U         U  !W         "W  SX         TX  Y         Y  Z         Z  [         [  ]         ]  M^         N^  _         _  `         `  a         a  c         c  Gd         Hd  ye         ze  f         f  g         g  i         i  Aj         Bj  sk         tk  l         l  m         Pp  q         q  r         r  s         s  u         u  Hv         Jv  zw         |w  x         x  y         y  {         {  B|         D|  t}         v}  ~         ~             
           <         >  n         p             ҅         ԅ             6         8  h         j             ̋         ΋              0         2  b         d             Ƒ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               5!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                          
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -A-    A-S;-x    S;-I-|    I-;}-\    ;}-!9-    !9--h    -b-    b-	n-	    	n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               F!         X!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             *         +  U         V                                  r  e                                              .  !          ]  C          |  b                                                          8         9  c          d   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ;         <  w         x                        +         ,  g         h             	         	             W         X                                   G         H                                   7         8  s         t                        '         (  c         d                                     S!         T!  "         "  #         #  %         %  B&         D&  ~'         '  (         (  )         )  2+         4+  n,         p,  -         -  .         .  "0         $0  ^1         `1  2         2  3         3  5         5  N6         P6  7         7  8         8  :         :  >;         @;  z<         |<  =         =  >         >  .@         0@  jA         lA  B         B  C         C  E          E  ZF         \F  G          :N  CO          }O  P          P  Q          R  S          FS  OT         m  o         o  Np           ו         ؕ             O         P             ǚ         Ț             ?         @  {         |                        /         0  k         l                                    [         \             Ӫ         Ԫ             K         L             ï         į              ;         <  w         x                        +         ,  g         h                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               M!         f!   2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     *
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-;-x    ;-/J-|    /J-}}-\    }}-I-    I--h    -ub-    ub-߂-	    ߂--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               B!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 	         	  F
         G
  q                                            D         E  o          p  i                                  !            \  U            }                                      !          "  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            0         2  b         d                                   *         ,  \         ^  	         	  
         
             $         &  V         X                                               P         R                                              J         L  |         ~                                   D          F   v!         x!  "         "  #          *  +          ,  ,          .-  .          X.  H/          /  r0         G  H         H  I         I  ,K         .K  ^L         `L  M         M  N         N  O         O  &Q         (Q  XR         ZR  S         S  T         T  U         U   W         "W  RX         TX  Y         Y  Z         Z  [         [  ]         ]  L^         N^  ~_         _  `         `  a         a  c         c  Fd         Hd  xe         ze  f         f  g         g  i         i  @j         Bj  rk         tk  l         l  m         Pp  q         q  r         r  s         s  u         u  Iv         Jv  {w         |w  x         x  y         y  {         {  C|         D|  u}         v}  ~         ~                        =         >  o         p             Ӆ         ԅ             7         8  i         j             ͋         ΋              1         2  c         d             Ǒ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               5!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                          
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -;-    ;-[6-x    [6-D-|    D-Cx-\    Cx--    --h    -b-    b-n-	    n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               @!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     >  #          ]  B          |  a                                    	          L	  M
          
                                   B         C  m         n                        #         $  U         V                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            8         :  r         t                                  "  Z         \             	         	           
  B         D  |         ~                        *         ,  d         f                                   L         N                                   4         6  n         p                        !         !  V"         X"  #         #  $          0  1          2  3          <3  64          p4  j5          5  6         H  J         J  WK         XK  L         L  M         M  O         O  ?P         @P  yQ         zQ  R         R  S         S  'U         (U  aV         bV  W         W  X         X  Z         Z  I[         J[  \         \  ]         ]  ^         ^  1`         2`  ka         la  b         b  c         c  e         e  Sf         Tf  g         g  h         h  j         j  ;k         <k  ul         vl  m         No  p         p  q         q  r         r  5t         6t  ou         pu  v         v  w         w  y         y  Wz         Xz  {         {  |         |  ~         ~  ?         @  y         z                        '         (  a         b             Շ         և             I         J                                   1         2  k         l             ߒ                              >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               @!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     +
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -G-    G-};-x    };-J-|    J-e}-\    e}--    --h    -cb-    cb-傳-	    傳--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               J!         V!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             *         +  U         V                                                                                   >  #          ]  V                                                           0         1  [         \               !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            2         4  f         h                                   6         8  j         l  	         	  
         
             :         <  n         p                        
           >         @  r         t                                   B         D  v         x                                   F         H  z          |   !         !  "         "  $         $  %          &  3'         4'  g(         h(  )         )  *         *  ,         ,  7-         8-  k.         l.  /         /  0         0  2         2  ;3         <3  o4         p4  5         5  6         6  8         8  ?9         @9  s:         t:  ;         ;  <         <  >         >  C?         D?  w@         x@  A         A  B         B  D         D  GE         HE  {F         |F  G         G  H          (U  (V          bV  bW          W  X          X  Y          Z  [         "o  Tp         Vp  q         q  r         r  s         s  $u         &u  Xv         Zv  w         w  x         x  y         y  ({         *{  \|         ^|  }         }  ~         ~             ,         .  `         b             Ȅ         ʄ             0         2  d         f             ̊         Ί              4         6  h         j             А         Ґ             8                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               .!         6!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                          
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -;-    ;-U6-x    U6-D-|    D-=x-\    =x--    --h    -b-    b-n-	    n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               B!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     >  $          ]  C          |  b                                    	          L	  N
          
                                   B         C  m         n                        !         "  R         S                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            9         :  s         t                        !         "  [         \             	         	  	         
  C         D  }         ~                        +         ,  e         f                                   M         N                                   5         6  o         p                        !         !  W"         X"  #         #  $          6  7          8  9          @9  ::          t:  n;          ;  <         H  J         J  VK         XK  L         L  M         M  O         O  >P         @P  xQ         zQ  R         R  S         S  &U         (U  `V         bV  W         W  X         X  Z         Z  H[         J[  \         \  ]         ]  ^         ^  0`         2`  ja         la  b         b  c         c  e         e  Rf         Tf  g         g  h         h   j         j  :k         <k  tl         vl  m         No  p         p  q         q  r         r  4t         6t  nu         pu  v         v  w         w  y         y  Vz         Xz  {         {  |         |  ~         ~  >         @  x         z                        &         (  `         b             ԇ         և             H         J                                   0         2  j         l             ޒ                              >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               B!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     *
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -A-    A-w;-x    w;-J-|    J-_}-\    _}--    --h    -ib-    ib-傳-	    傳--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               F!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             *         +  U         V                                                                                   >  $          ]  U                                                           +         ,  V         W               !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            3         4  g         h                                   7         8  k         l  	         	  
         
             ;         <  o         p                                   ?         @  s         t                                   C         D  w         x                                   G         H  {          |   !         !  "         "  $         $  %          &  2'         4'  f(         h(  )         )  *         *  ,         ,  6-         8-  j.         l.  /         /  0         0  2         2  :3         <3  n4         p4  5         5  6         6  
8         8  >9         @9  r:         t:  ;         ;  <         <  >         >  B?         D?  v@         x@  A         A  B         B  D         D  FE         HE  zF         |F  G         G  H          J[  J\          \  ]          ]  ^          ^  _          2`  2a         "o  Up         Vp  q         q  r         r  s         s  %u         &u  Yv         Zv  w         w  x         x  y         y  ){         *{  ]|         ^|  }         }  ~         ~             -         .  a         b             Ʉ         ʄ             1         2  e         f             ͊         Ί             5         6  i         j             ѐ         Ґ             9                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               0!         :!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                          
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     ---X     ---     -Y-    Y-;-x    ;-I-|    I-|-\    |-
-    
--h    -b-    b-qx-	    qx--h
    -K׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               S!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      @         A  k	         l	  
         
           I  s         t                                                          B            d  <            ^                                              (             W   @!          !  o"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            7         8  o         p                                   O         P             	         	  
         
  /         0  g         h                                   G         H                                   '         (  _         `                                   ?         @  w         x                          "          "  W#         X#  $         $  %         %  &          '  6(         8(  n)         p)  *         *  +         +  -         -  N.         P.  /         /  0         0  1         1  .3         03  f4         h4  5         5  6         6  8         8  F9         H9  ~:         :  ;         ;  <         <  &>         (>  ^?         `?  @         @  A         A  C         C  >D         @D  vE         xE  F         F  G         G  I         I  K         K  ML         NL  M         M  N         N  O         O  -Q         .Q  eR         fR  S         S  T         T  V         V  EW         FW  }X         ~X  Y         Y  Z         Z  %\         &\  ]]         ^]  ^         ^  _         _  a         a  =b         >b  uc         vc  d         d  e         e  g         g  Uh         Vh  i         i  j         j  k         k  5m         6m  mn            |          Ɔ              ̈                      >                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               P!         V!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 e
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -M-    M-;-x    ;-;J-|    ;J-}-\    }-2-    2--P    -b-    b-c-	    c-C-D
    C-K׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               N!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 	         	  F
         G
  q                                            D         E  o          p  a                                  !            \  M            |                                      !          "  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            5         6  k         l                                   C         D  y         z  	         	  
         
             Q         R                                   )         *  _         `                                   7         8  m         n                                   E         F  {         |               !         !  #         #  S$         T$  %         %  &         &  '         '  +)         ,)  a*         b*  +         +  ,         ,  .         .  9/         :/  o0         p0  1         1  2         2  4         4  G5         H5  }6         ~6  7         7  8         8  :          :  U;         V;  <         <  =         =  >         >  -@         .@  cA         dA  B         B  C         C  E         E  ;F         <F  pG         rG  H         H  I         nn  o         o  p         p  r         r  Ds         Fs  zt         |t  u         u  v         v  x         x  Ry         Ty  z         z  {         {  |         |  *~         ,~  `         b             ́         ΁             8         :  n         p             ڇ         ܇             F         H  |         ~                       v              ®                      ?  H                               >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               F!         T!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 i
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -M-    M-;-x    ;-I-|    I-|-\    |-q-    q--h    -b-    b-qx-	    qx--h
    -K׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               O!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      @         A  k	         l	  
         
           I  s         t                                                          A            b  :            [            |                                  #             R   :!          !  i"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            6         8  n         p                                   N         P             	         	  
         
  .         0  f         h                                   F         H  ~                                 &         (  ^         `                                   >         @  v         x                          "          "  V#         X#  $         $  %         %  &          '  7(         8(  o)         p)  *         *  +         +  -         -  O.         P.  /         /  0         0  1         1  /3         03  g4         h4  5         5  6         6  8         8  G9         H9  :         :  ;         ;  <         <  '>         (>  _?         `?  @         @  A         A  C         C  ?D         @D  wE         xE  F         F  G         G  I         I  K         K  LL         NL  M         M  N         N  O         O  ,Q         .Q  dR         fR  S         S  T         T  V         V  DW         FW  |X         ~X  Y         Y  Z         Z  $\         &\  \]         ^]  ^         ^  _         _  a         a  <b         >b  tc         vc  d         d  e         e  g         g  Th         Vh  i         i  j         j  k         k  4m         6m  ln          f  D            l                      ގ                                 >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               R!         Z!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 e
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -M-    M-;-x    ;-;J-|    ;J-}-\    }-g8-    g8--P    -b-    b-c-	    c-C-D
    C-K׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               M!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 	         	  F
         G
  q                                            D         E  o          p  b                                  !            \  N            }                                      !          "  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            4         6  j         l                                   B         D  x         z  	         	  
         
             P         R                                   (         *  ^         `                                    6         8  l         n                                   D         F  z         |               !         !  #         #  R$         T$  %         %  &         &  '         '  *)         ,)  `*         b*  +         +  ,         ,  .         .  8/         :/  n0         p0  1         1  2         2  4         4  F5         H5  |6         ~6  7         7  8         8  :          :  T;         V;  <         <  =         =  >         >  ,@         .@  bA         dA  B         B  C         C  E         E  :F         <F  qG         rG  H         H  I         nn  o         o  p         p  r         r  Es         Fs  {t         |t  u         u  v         v  x         x  Sy         Ty  z         z  {         {  |         |  +~         ,~  a         b             ́         ΁             9         :  o         p             ۇ         ܇             G         H  }         ~                       Ų  γ                      K  T                      ѷ  ڸ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               H!         T!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 i
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -S-    S-=6-x    =6-D-|    D-%x-\    %x-7-    7--h    -b-    b-}x-	    }x--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               S!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     >  #          ]  B          |  a                                    	          L	  H
          
                                   B         C  m         n                                   @         A  k         l                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                        4  (          h  \                                 U$  %         %  &         &  '         '  ,)         -)  b*         c*  +         +  ,         ,  .         .  :/         ;/  p0         q0  1         1  2         2  4         4  H5         I5  ~6         6  7         7  8         8   :         !:  V;         W;  <         <  =         =  >         >  .@         /@  dA         eA  B         B  C         C  E         E  <F         =F  rG         G  I         I  CJ         EJ  yK         {K  L         L  M         M  O         O  QP         SP  Q         Q  R         R  S         S  )U         +U  _V         aV  W         W  X         X  Z         Z  7[         9[  m\         o\  ]         ]  ^         ^  `         `  Ea         Ga  {b         }b  c         c  d         d  f         f  Sg         Ug  h         h  i         i  j         j  +l         l  
n         ړ             E         F  {         |                                   S         T                                   +         ,  a         b             ͤ         Τ             9         :  o         p             ۪         ܪ             G         H  }         ~                                    U         V                      µ             -                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               G!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 e
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -A-    A-_;-    _;-I-    I-G}-d    G}--    --p    -b-    b-?-
    ?-M-
    M-۰-t    ۰-E׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               I!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             *         +  U         V                                                                                   >  #          ]  S                                                                      8         9  c          d   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                        &            L  ;          r  a                     T#  $         $  %         %  &         &  3(         4(  k)         l)  *         *  +         +  -         -  K.         L.  /         /  0         0  1         1  +3         ,3  c4         d4  5         5  6         6  8         8  C9         D9  {:         |:  ;         ;  <         <  #>         $>  [?         \?  @         @  A         A  C         C  ;D         <D  sE         tE  F         F  G         G  I         I  RJ         TJ  K         K  L         L  M         M  2O         4O  jP         lP  Q         Q  R         R  T         T  JU         LU  V         V  W         W  X         X  *Z         ,Z  b[         d[  \         \  ]         ]  
_         _  B`         D`  za         |a  b         b  c         c  "e         $e  Zf         \f  g         g  h         h  j         j  :k         <k  rl         tl  m         m  n         n  p         p  Rq         Tq  r         r  s         s  t         t  2v         4v  jw         lw  x         x  y         y  {         {  J|         L|  }         }  ~         ~             *         ,  b         d             ҄         Ԅ  
           B         D  z         |                        "         $  Z         \             ʏ         ̏                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               Q!         b!  2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    r
         s
             X      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     &
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -M-    M-=6-x    =6-D-|    D-%x-\    %x-7-    7--h    -b-    b-}x-	    }x--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               O!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     >  $          ]  C          |  b                                    	          L	  G
          
                                   B         C  m         n                                   @         A  k         l                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                     8  ,          l  `	          	  
          
           U$  %         %  &         &  '         '  +)         -)  a*         c*  +         +  ,         ,  .         .  9/         ;/  o0         q0  1         1  2         2  4         4  G5         I5  }6         6  7         7  8         8  :         !:  U;         W;  <         <  =         =  >         >  -@         /@  cA         eA  B         B  C         C  E         E  ;F         =F  qG         G  I         I  DJ         EJ  zK         {K  L         L  M         M  O         O  RP         SP  Q         Q  R         R  S         S  *U         +U  `V         aV  W         W  X         X  Z         Z  8[         9[  n\         o\  ]         ]  ^         ^  `         `  Fa         Ga  |b         }b  c         c  d         d  f         f  Tg         Ug  h         h  i         i  j         j  ,l         l  	n         ړ             D         F  z         |                                   R         T                                   *         ,  `         b             ̤         Τ             8         :  n         p             ڪ         ܪ             F         H  |         ~                                    T         V                      µ             ,                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               G!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 e
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -A-    A-_;-    _;-I-    I-G}-d    G}--    --p    -b-    b-?-
    ?-M-
    M-հ-t    հ-E׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               I!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             *         +  U         V                                                                                   >  $          ]  R                                                                      8         9  c          d   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                 
            0	  
          V
  E         T#  $         $  %         %  &         &  2(         4(  j)         l)  *         *  +         +  -         -  J.         L.  /         /  0         0  1         1  *3         ,3  b4         d4  5         5  6         6  
8         8  B9         D9  z:         |:  ;         ;  <         <  ">         $>  Z?         \?  @         @  A         A  C         C  :D         <D  rE         tE  F         F  G         G  I         I  SJ         TJ  K         K  L         L  M         M  3O         4O  kP         lP  Q         Q  R         R  T         T  KU         LU  V         V  W         W  X         X  +Z         ,Z  c[         d[  \         \  ]         ]  _         _  C`         D`  {a         |a  b         b  c         c  #e         $e  [f         \f  g         g  h         h  j         j  ;k         <k  sl         tl  m         m  n         n  p         p  Sq         Tq  r         r  s         s  t         t  3v         4v  kw         lw  x         x  y         y  {         {  K|         L|  }         }  ~         ~             +         ,  c         d             ӄ         Ԅ             C         D  {         |                        #         $  [         \             ˏ         ̏                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               Q!         b!  2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    r
         s
             X      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     %
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-:-x    :-I-|    I-|-\    |-A-    A--h    -b-    b--	    --h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               A!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      @         A  k	         l	  
         
           I  s         t                                                          ;            V  :          q  U            p                                               4   0!          c!  _"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            2         4  f         h                                   6         8  j         l  	         	  
         
             :         <  n         p                        
           >         @  r         t                                   B         D  v         x                                   F         H  z          |   !         !  "         "  $         $  J%         L%  ~&         &  '         '  (         (  *         *  N+         P+  ,         ,  -         -  .         .  0          0  R1         T1  2         2  3         3  4         4  "6         $6  V7         X7  8         8  9         9  :         :  &<         (<  Z=         \=  >         >  ?         ?  @         @  *B         ,B  ^C         `C  D         D  E         E  F         F  .H         0H  bI         dI  J         J  K         K  L          M  2N         4N  fO         hO  P         P  Q         Q  S         S  6T         8T  jU         lU  V         W  QX         SX  Y         Y  Z         Z  [         [  !]         #]  U^         W^  _         _  `         `  a         a  %c         'c  Yd         [d  e         e  f         f  g         g  )i         +i  ]j          lw  px          x  y          y  z          {  |          L|  P}         T                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               @!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   2      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     -;-    ;-G5-    G5-C-    C-/w-l    /w-O2-    O2--x    -b-	    b-9-
    9-G-
    G--|    -ճ-    ճ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               G!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         /  %          ^  T                                                          !            <	  
          W
  5         r                                              H                      %         &  P         Q  {         |                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |        "  #         #  !%         #%  [&         ]&  '         '  (         (  	*         *  C+         E+  },         ,  -         -  .         .  +0         -0  e1         g1  2         2  3         3  5         5  M6         O6  7         7  8         8  9         9  5;         7;  o<         q<  =         =  >         >  @         @  WA         YA  B         B  C         C  E         E  ?F          Q  gR          R  S          S  T          T  U          V  V         h  8i         :i  rj         tj  k         k  l         l   n         "n  Zo         \o  p         p  q         q  s         
s  Bt         Dt  |u         ~u  v         v  w         w  *y         ,y  dz         fz  {         {  |         |  ~         ~  L         N                                   4         6  n         p                                   V         X             ʌ         ̌             =         ?  w         y                        #         %  ]         _             ї         ӗ             E         G                                   -         /  g         i             ۢ         ݢ             O         Q             ç         ŧ             7         9  q         s                                 !  Y         [             Ͳ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               F!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    r
         s
             X      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    r
         s
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     y
            0     -,-X     ,--     -5-    5-:-x    :-I-|    I-|-\    |-A-    A--h    -b-    b--	    --h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               C!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      @         A  k	         l	  
         
           I  s         t                                                          ;             V  ;          q  V            q                                                4   /!          c!  ^"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            3         4  g         h                                   7         8  k         l  	         	  
         
             ;         <  o         p                                   ?         @  s         t                                   C         D  w         x                                   G         H  {          |   !         !  "         "  $         $  K%         L%  &         &  '         '  (         (  *         *  O+         P+  ,         ,  -         -  .         .  0          0  S1         T1  2         2  3         3  4         4  #6         $6  W7         X7  8         8  9         9  :         :  '<         (<  [=         \=  >         >  ?         ?  @         @  +B         ,B  _C         `C  D         D  E         E  F         F  /H         0H  cI         dI  J         J  K         K  L          M  3N         4N  gO         hO  P         P  Q         Q  S         S  7T         8T  kU         lU  V         W  RX         SX  Y         Y  Z         Z  [         [  "]         #]  V^         W^  _         _  `         `  a         a  &c         'c  Zd         [d  e         e  f         f  g         g  *i         +i  ^j          }  ~          ~                        ,  0          d  h         T                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               <!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   1      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     -;-    ;-G5-    G5-C-    C-/w-l    /w-U2-    U2--x    -b-	    b-9-
    9-G-
    G--|    -ճ-    ճ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               G!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         /  $          ^  S                                                          !   	          <	  
          W
  6         r                                              H                      %         &  P         Q  {         |                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |        "  #         #  "%         #%  \&         ]&  '         '  (         (  
*         *  D+         E+  ~,         ,  -         -  .         .  ,0         -0  f1         g1  2         2  3         3  5         5  N6         O6  7         7  8         8  9         9  6;         7;  p<         q<  =         =  >         >  @         @  XA         YA  B         B  C         C  E         E  @F          !W  X          AX  'Y          aY  GZ          Z  g[          [  \         h  9i         :i  sj         tj  k         k  l         l  !n         "n  [o         \o  p         p  q         q  	s         
s  Ct         Dt  }u         ~u  v         v  w         w  +y         ,y  ez         fz  {         {  |         |  ~         ~  M         N                                   5         6  o         p                                   W         X             ˌ         ̌             >         ?  x         y                        $         %  ^         _             җ         ӗ             F         G                                   .         /  h         i             ܢ         ݢ             P         Q             ħ         ŧ             8         9  r         s                                  !  Z         [             β                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               D!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    r
         s
             X      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    r
         s
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     z
            0     -,-X     ,--     -G-    G-S;-x    S;-I-|    I-;}-\    ;}--    --h    -ob-    ob--	    --h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               O!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             *         +  U         V                                  r  n                                              .  *          ]  B          |  a                                                          8         9  c          d   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            0         2  b         d                                   *         ,  \         ^  	         	  
         
             $         &  V         X                                               P         R                                              J         L  |         ~                                   D          F   v!         x!  "         "  #          4;  (<          \<  P=          =  x>          >  ?          ?  @         F  G         G  &I         (I  XJ         ZJ  K         K  L         L  M         M   O         "O  RP         TP  Q         Q  R         R  S         S  U         U  LV         NV  ~W         W  X         X  Y         Y  [         [  F\         H\  x]         z]  ^         ^  _         _  a         a  @b         Bb  rc         tc  d         d  e         e  g         
g  :h         <h  li         ni  j         j  k         k  m         m  5n         6n  go         ho  p         p  q         q  r         r  /t         0t  au         bu  v         v  w         w  x         x  )z         *z  [{         \{  |         |  }         }  ~         ~  #         $  U         V                                              O         P                                              I         J  {                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               7!         >!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   2      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -A-    A-S;-x    S;-I-|    I-;}-\    ;}--    --h    -b-    b--n-	    -n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               F!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             *         +  U         V                                                                                   >  #          ]  S                                                                      8         9  c          d   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ;         <  w         x                        +         ,  g         h             	         	             W         X                        
           F         H                                   6         8  r         t                        &         (  b         d                                     R!         T!  "         "  #         #  %          =  >          >  ?          ?  @          A  B          JB  @C         :H  uI         vI  J         J  K         K  )M         *M  eN         fN  O         O  P         P  R         R  US         VS  T         T  U         U  	W         
W  EX         FX  Y         Y  Z         Z  [         [  5]         6]  q^         r^  _         _  `         `  %b         &b  ac         bc  d         d  e         e  g         g  Qh         Rh  i         i  j         j  l         l  Am         Bm  |n         ~n  o         o  p         p  0r         2r  ls         ns  t         t  u         u   w         "w  \x         ^x  y         y  z         z  |         |  L}         N}  ~         ~                         <         >  x         z                        ,         .  h         j                                   X         Z             Џ         ҏ             H                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               G!         Z!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     O
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -A-    A-S;-x    S;-I-|    I-;}-\    ;}--    --h    -ob-    ob--	    --h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               M!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             *         +  U         V                                  r  m                                              .  )          ]  C          |  b                                                          8         9  c          d   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            1         2  c         d                                   +         ,  ]         ^  	         	  
         
             %         &  W         X                                               Q         R                                              K         L  }         ~                                   E          F   w!         x!  "         "  #          @  A          $B  C          LC  @D          tD  hE          E  F         F  G         G  'I         (I  YJ         ZJ  K         K  L         L  M         M  !O         "O  SP         TP  Q         Q  R         R  S         S  U         U  MV         NV  W         W  X         X  Y         Y  [         [  G\         H\  y]         z]  ^         ^  _         _  a         a  Ab         Bb  sc         tc  d         d  e         e  	g         
g  ;h         <h  mi         ni  j         j  k         k  m         m  4n         6n  fo         ho  p         p  q         q  r         r  .t         0t  `u         bu  v         v  w         w  x         x  (z         *z  Z{         \{  |         |  }         }  ~         ~  "         $  T         V                                              N         P                                              H         J  z                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               5!         >!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   1      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -u -X     u -ְ-     ְ-=ܰ-    =ܰ-c4-p    c4-!@-T    !@-h-4    h--    --8    -b-    b--n-	    -n--8
    -?׳-
                   =         b      J      7   U   7   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               H!         T!  V1                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 64 })  outputElements  5184    input0Elements  31104   input1Elements  24576   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     4                           #         b      J      7   M     operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       @             *         +  U         V                       h	  N
          
  m                                                          B  7          q  f           F         G  q         r                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     4                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  1728    input0Elements  1728    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       x            :         <  v         x                        *         ,  f         h             	         	             V         X                                   G         H                                   7         8  s         t                        '         (  c         d                                     S!         T!  "         "  #         #  %          zC  pD          D  E          E  F          
G   H         :H  tI         vI  J         J  K         K  (M         *M  dN         fN  O         O  P         P  R         R  TS         VS  T         T  U         U  W         
W  DX         FX  Y         Y  Z         Z  [         [  4]         6]  p^         r^  _         _  `         `  $b         &b  `c         bc  d         d  e         e  g         g  Ph         Rh  i         i  j         j  l         l  @m         Bm  }n         ~n  o         o  p         p  1r         2r  ms         ns  t         t  u         u  !w         "w  ]x         ^x  y         y  z         z  |         |  M}         N}  ~         ~                        =         >  y         z                        -         .  i         j                                   Y         Z             я         ҏ             I                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               G!         Z!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     N
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -A-    A-:-    :-I-    I-|-d    |-
-    
-W-0    W--    -b-H	    b-c-P
    c-[-
    [-9׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               J!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      @         A  k	         l	  
         
           I  s         t                                                          ;            V  :          q  U            p                                               4   0!          c!  _"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q   w  operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       \            .         0  ^         `                                               N         P  ~	         	  
         
                        >         @  n         p                                    .         0  ^         `                                               N         P  ~                                              >!         @!  n"         %  &         &  ((         *(  X)         Z)  *         *  +         +  ,         ,  .         .  H/         J/  x0         z0  1         1  2         2  4         
4  85         :5  h6         j6  7         7  8         8  9         9  (;         *;  X<         Z<  =         =  >         >  ?         ?  A         A  HB         JB  xC         zC  D         D  E         E  G         
G  8H          h  i          j  j          2k  +l          bl  [m          m  n         n  o         o   q         "q  Pr         Rr  s         s  t         t  u         u  w         w  @x         Bx  py         ry  z         z  {         {   }         }  0~         2~  `         b                                              L         N  {         }             ۈ         ݈             ;                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               F!         \!  1         
2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 u
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -;-    ;-A5-x    A5-C-|    C-)w-\    )w--    --h    -b-    b-x-	    x--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               F!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         /  %          ^  T                                                          !            <	  
          W
  5         r                                              H                      %         &  P         Q  {         |                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       @  C                                              &  '         '  (         (  
*         *  6+         7+  b,         c,  -         -  .         .  /         /  1         1  >2         ?2  j3         k3  4         4  5         5  6         6  8         8  F9         G9  r:         s:  ;         ;  <         <  =         =  !?         #?  M@         O@  yA         {A  B         B  C         C  D         D  )F         +F  UG         WG  H         H  I         I  J         J  L         L  1M         3M  ]N         _N  O         O  P         P  Q         Q  S         S  9T         ;T  eU         gU  V         V  W         W  X         X  Z         Z  A[         C[  m\         o\  ]         ]  ^         ^  _         _  a         a  Ib         Kb  uc         wc  d         d  e         e  f         f  %h         'h  Qi         Si  }j         j  k         k  l          n  *o         ,o  Vp         Xp  q         q  r         r  s         s  u         u  2v         4v  ^w         `w  x         x  y         y  z         z  |         |  :}         <}  f~         h~                                              B         D  n         p             Ƈ         ȇ                         J         L  v         x             Ύ         Ў             &                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               F!         X!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 y
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -;-    ;-:-    :-I-    I-|-d    |-
-    
-W-0    W--    -b-H	    b-c-P
    c-[-
    [-9׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               G!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      @         A  k	         l	  
         
           I  s         t                                                          ;             V  ;          q  V            q                                                4   /!          c!  ^"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q   w  operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       \            /         0  _         `                                               O         P  	         	  
         
                        ?         @  o         p                                    /         0  _         `                                               O         P                                                ?!         @!  o"         %  &         &  )(         *(  Y)         Z)  *         *  +         +  ,         ,  .         .  I/         J/  y0         z0  1         1  2         2  	4         
4  95         :5  i6         j6  7         7  8         8  9         9  );         *;  Y<         Z<  =         =  >         >  ?         ?  A         A  IB         JB  yC         zC  D         D  E         E  	G         
G  9H         h  j         j  1k         2k  al         bl  m         m  n          n  o          o  p          "q  r          Rr  Ks          s  {t         t  u         u  w         w  Ax         Bx  qy         ry  z         z  {         {  }         }  1~         2~  a         b                                              M         N  |         }             ܈         ݈             <                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               B!         X!  1         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 u
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -;-    ;-A5-x    A5-C-|    C-)w-\    )w--    --h    -b-    b-x-	    x--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               H!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         /  $          ^  S                                                          !   	          <	  
          W
  6         r                                              H                      %         &  P         Q  {         |                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         @  C                        	           
            @  C         &  '         '  (         (  	*         *  5+         7+  a,         c,  -         -  .         .  /         /  1         1  =2         ?2  i3         k3  4         4  5         5  6         6  8         8  E9         G9  q:         s:  ;         ;  <         <  =         =  "?         #?  N@         O@  zA         {A  B         B  C         C  D         D  *F         +F  VG         WG  H         H  I         I  J         J  L         L  2M         3M  ^N         _N  O         O  P         P  Q         Q  S         S  :T         ;T  fU         gU  V         V  W         W  X         X  Z         Z  B[         C[  n\         o\  ]         ]  ^         ^  _         _  a         a  Jb         Kb  vc         wc  d         d  e         e  f         f  &h         'h  Ri         Si  ~j         j  k         k  l          n  +o         ,o  Wp         Xp  q         q  r         r  s         s  u         u  3v         4v  _w         `w  x         x  y         y  z         z  |         |  ;}         <}  g~         h~                                              C         D  o         p             Ǉ         ȇ                         K         L  w         x             ώ         Ў             '                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               D!         T!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 y
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -/-    /-M;-    M;-I-    I-5}-d    5}-!-    !--p    -b-    b---
    --;-
    ;--t    -3׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               E!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             *         +  U         V                                  r  n                                              .  *          ]  B          |  a                                                          8         9  c          d   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                    	          @  I                     %  &         &  '         '  )         
)  6*         8*  d+         f+  ,         ,  -         -  .         .  0         0  J1         L1  x2         z2  3         3  4         4  6         6  07         27  ^8         `8  9         9  :         :  ;         ;  =         =  D>         F>  r?         t?  @         @  A         A  B         B  *D         ,D  XE         ZE  F         F  G         G  H         H  J         J  >K         @K  lL         nL  M         M  N         N  O         O  %Q         &Q  SR         TR  S         S  T         T  U         U  W         W  9X         :X  gY         hY  Z         Z  [         [  \         \  ^          ^  M_         N_  {`         |`  a         a  b         b  d         d  3e         4e  af         bf  g         g  h         h  i         i  k         k  Gl         Hl  um         vm  n         n  o         o  p          q  ,r         .r  Zs         \s  t         t  u         u  v         v  x         x  @y         By  nz         pz  {         {  |         |  }         }  &         (  T         V                        ރ                      :         <  h         j             ĉ         Ɖ                       "  N         P  |         ~                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               O!         f!  2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    r
         s
             X      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     P
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -A-    A-;-x    ;-/J-|    /J-}}-\    }}-,-    ,--h    -Qb-    Qb-ق-	    ق--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               H!         \!  1         
2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 %	         &	  P
         Q
  {                                 #         $  N         O  y          z  |                                  +  -          f  h                                                  !          "  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            -         .  [         \                                              A         B  o	         p	  
         
                        '         (  U         V                                              ;         <  i         j                                   !         "  O         P  }         ~                        !         !  5"         6"  c#         %  &         &  '         '  %)         &)  S*         T*  +         +  ,         ,  -         -  /         /  90         :0  g1         h1  2         2  3         3  4         4  6          6  M7         N7  {8         |8  9         9  :         :  <         <  3=         4=  a>         b>  ?         ?  @         @  A         A  C         C  GD         HD  uE         vE  F         F  G         G  H         TI  J         J  K          u  v          v  w          x  x          7y  z          Sz  5{           ?         A  m         o             ɐ         ː             %         '  S         U                        ݗ         ߗ             9         ;  g         i             Ý         ŝ                      !  M         O  {         }             פ         ٤             3         5  a         c                                                      >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               )!         8!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                          
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -/-    /-M;-    M;-I-    I-5}-d    5}-!-    !--p    -b-    b---
    --;-
    ;--t    -3׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               C!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             *         +  U         V                                  r  m                                              .  )          ]  C          |  b                                                          8         9  c          d   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                        	          @  I                                 %  &         &  '         '  	)         
)  7*         8*  e+         f+  ,         ,  -         -  .         .  0         0  K1         L1  y2         z2  3         3  4         4  6         6  17         27  _8         `8  9         9  :         :  ;         ;  =         =  E>         F>  s?         t?  @         @  A         A  B         B  +D         ,D  YE         ZE  F         F  G         G  H         H  J         J  ?K         @K  mL         nL  M         M  N         N  O         O  $Q         &Q  RR         TR  S         S  T         T  U         U  
W         W  8X         :X  fY         hY  Z         Z  [         [  \         \  ^          ^  L_         N_  z`         |`  a         a  b         b  d         d  2e         4e  `f         bf  g         g  h         h  i         i  k         k  Fl         Hl  tm         vm  n         n  o         o  p          q  -r         .r  [s         \s  t         t  u         u  v         v  x         x  Ay         By  oz         pz  {         {  |         |  }         }  '         (  U         V                        ߃                      ;         <  i         j             ŉ         Ɖ             !         "  O         P  }         ~                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               O!         f!  2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    r
         s
             X      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     Q
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -;-    ;-;-x    ;-5J-|    5J-}-\    }-,-    ,--h    -Wb-    Wb-ق-	    ق--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               F!         X!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                  *	         +	  U
         V
                                   (         )  S         T  ~                                              0  1          k  l                                               !  !          ""  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ,         .  Z         \                                              @         B  n	         p	  
         
                        &         (  T         V                                              :         <  h         j                                             "  N         P  |         ~                        !         !  4"         6"  b#         %  &         &  '         '  $)         &)  R*         T*  +         +  ,         ,  -         -  
/         /  80         :0  f1         h1  2         2  3         3  4         4  6          6  L7         N7  z8         |8  9         9  :         :  <         <  2=         4=  `>         b>  ?         ?  @         @  A         A  C         C  FD         HD  tE         vE  F         F  G         G  H         TI  J         J  K          o{  Q|          |  m}          }  ~          ~                         @         A  n         o             ʐ         ː             &         '  T         U                        ޗ         ߗ             :         ;  h         i             ĝ         ŝ                       !  N         O  |         }             ؤ         ٤             4         5  b         c                                                      >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               +!         <!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                          
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     -/-    /-%6-    %6-D-    D-x-l    x-4-    4--x    -b-	    b---
    --;-
    ;--|    -ճ-    ճ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               C!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     >  #          ]  B          |  a                                    	          L	  H
          
                                   B         C  m         n                                   @         A  k         l                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |        D%  n&          ?  @          @  A          2B  ;C          uC  ~D          D  E         JL  uM         vM  N         N  O         O  P         P  %R         &R  QS         RS  }T         ~T  U         U  V         V  X         X  -Y         .Y  YZ         ZZ  [         [  \         \  ]         ]  	_         
_  5`         6`  aa         ba  b         b  c         c  d         d  f         f  =g         >g  ih         jh  i         i  j         j  k         k  m         m  En         Fn  qo         ro  p         p  q         q  r         r  !t         "t  Mu         Nu  yv         zv  w         w  x         x  y         y  ){         *{  U|         V|  }         }  ~         ~                        1         2  ]         ^                                              9         :  e         f                                              A         B  m         n             ē         Ɠ                        H         J  t         v             ̚         Κ             $         &  P         R  |         ~             ԡ         ֡              ,         .  X         Z                        ܨ         ި           
  4         6  `         b                                              <         >  h         j                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               D!         T!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    r
         s
             X      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    r
         s
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            0     -,-X     ,--     -#-    #-G;-x    G;-I-|    I-/}-\    /}-{-    {--h    -?b-    ?b-ӂ-	    ӂ--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               :!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             *         +  U         V                                                                                   >  #          ]  S                                                                      8         9  c          d   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            .         0  ^         `                                               N         P  ~	         	  
         
                        >         @  n         p                                    .         0  ^         `                                               N         P  ~                                              >!         @!  n"         p"  #         D%  s&         t&  '         '  (         (  *         *  3+         4+  c,         d,  -         -  .         .  /         /  #1         $1  S2         T2  3         3  4         4  5         5  7         7  C8         D8  s9         t9  :         :  ;         ;  =         =  3>         4>  c?         d?  @         @  A         A  B         B  #D         $D  SE         TE  F         F  G         G  H          c  d          d  e          f  g          >g  0h          jh  \i         ir  s         s  t         t  u         u  'w         )w  Wx         Yx  y         y  z         z  {         {  }         }  G~         I~  w         y             ׁ         ف           	  7         9  g         i             Ǉ         ɇ             '         )  W         Y                                              G         I  w         y             ה         ٔ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               $!         .!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                          
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     -)-    )-%6-    %6-D-    D-x-l    x-4-    4--x    -b-	    b---
    --;-
    ;--|    -ճ-    ճ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               A!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     >  $          ]  C          |  b                                    	          L	  G
          
                                   B         C  m         n                                   @         A  k         l                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |        D%  o&          E  G          >G  GH          H  I          I  J          K  L         JL  tM         vM  N         N  O         O  P         P  $R         &R  PS         RS  |T         ~T  U         U  V         V   X         X  ,Y         .Y  XZ         ZZ  [         [  \         \  ]         ]  _         
_  4`         6`  `a         ba  b         b  c         c  d         d  f         f  <g         >g  hh         jh  i         i  j         j  k         k  m         m  Dn         Fn  po         ro  p         p  q         q  r         r   t         "t  Lu         Nu  xv         zv  w         w  x         x  y         y  ({         *{  T|         V|  }         }  ~         ~                        0         2  \         ^                                              8         :  d         f                                              @         B  l         n             œ         Ɠ                        I         J  u         v             ͚         Κ             %         &  Q         R  }         ~             ա         ֡             -         .  Y         Z                        ݨ         ި  	         
  5         6  a         b                                              =         >  i         j                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               B!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    r
         s
             X      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    r
         s
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            0     -,-X     ,--     --    -G;-x    G;-I-|    I-/}-\    /}-{-    {--h    -Eb-    Eb-ӂ-	    ӂ--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               8!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             *         +  U         V                                                                                   >  $          ]  R                                                                      8         9  c          d   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            /         0  _         `                                               O         P  	         	  
         
                        ?         @  o         p                                    /         0  _         `                                               O         P                                                ?!         @!  o"         p"  #         D%  r&         t&  '         '  (         (  *         *  2+         4+  b,         d,  -         -  .         .  /         /  "1         $1  R2         T2  3         3  4         4  5         5  7         7  B8         D8  r9         t9  :         :  ;         ;  =         =  2>         4>  b?         d?  @         @  A         A  B         B  "D         $D  RE         TE  F         F  G         G  H          i  j          j  k          k  l          m  n          Fn  8o         ir  s         s  t         t  u         u  (w         )w  Xx         Yx  y         y  z         z  {         {  }         }  H~         I~  x         y             ؁         ف           	  8         9  h         i             ȇ         ɇ             (         )  X         Y                                              H         I  x         y             ؔ         ٔ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               &!         2!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                          
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -;-    ;-:-x    :-I-|    I-|-\    |--    --h    -b-    b-ex-	    ex--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               H!         \!  1         
2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      @         A  k	         l	  
         
           I  s         t                                                          ;            V  :          q  U            p                                               4   0!          c!  _"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            :         <  v         x                        *         ,  f         h             	         	             V         X                        
           F         H                                   6         8  r         t                        &         (  b         d                                     R!         T!  "         "  #         #  %         %  B&         D&  ~'         '  (         (  )         )  2+         4+  n,         p,  -         -  .         .  "0         $0  ^1         `1  2         2  3         3  5         5  N6         P6  7         7  8         8  :         :  >;         @;  z<         |<  =         =  >         >  .@         0@  jA         lA  B         B  C         C  E          E  ZF         \F  G         G  H         H  J         J  WK         XK  L         L  M         M  O         O  GP         HP  Q         Q  R         R  S         S  7U         8U  sV         tV  W         W  X         X  'Z         (Z  c[         d[  \         \  ]         ]  _         _  S`         T`  a         a  b         b  d         d  Ce         De  f         f  g         g  h         h  3j         4j  ok         pk  l         l  m         m  #o          ro  gp          p  q          q  r          r  s          "t  u                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               Y!         Z!  2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 e
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-;-x    ;-)J-|    )J-w}-\    w}--    --P    -b-    b-c-	    c-1-D
    1--׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               G!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 %	         &	  P
         Q
  {                                 #         $  N         O  y          z  |                                  +  -          f  h                                                  !          "  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            1         2  c         d                                   +         ,  ]         ^  	         	  
         
             %         &  W         X                                               Q         R                                              K         L  }         ~                                   E          F   w!         x!  "         "  #         #  %         %  ?&         @&  q'         r'  (         (  )         )  +         +  9,         :,  k-         l-  .         .  /         /  1         1  32         42  e3         f3  4         4  5         5  6         6  -8         .8  _9         `9  :         :  ;         ;  <         <  '>         (>  Y?         Z?  @         @  A         A  B         B  !D         "D  SE         TE  F         F  G         G  H         H  J          n  o          o  p          &q  /r         ir  s         s  t         t  u         u  /w         1w  ax         cx  y         y  z         z  {         {  )}         +}  [~         ]~                                   #         %  U         W                                              O         Q                                              I          	            L  U                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               K!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 i
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-:-x    :-I-|    I-|-\    |-c-    c--h    -b-    b-ex-	    ex--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               F!         X!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      @         A  k	         l	  
         
           I  s         t                                                          ;             V  ;          q  V            q                                                4   /!          c!  ^"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ;         <  w         x                        +         ,  g         h             	         	             W         X                                   G         H                                   7         8  s         t                        '         (  c         d                                     S!         T!  "         "  #         #  %         %  C&         D&  '         '  (         (  )         )  3+         4+  o,         p,  -         -  .         .  #0         $0  _1         `1  2         2  3         3  5         5  O6         P6  7         7  8         8  :         :  ?;         @;  {<         |<  =         =  >         >  /@         0@  kA         lA  B         B  C         C  E          E  [F         \F  G         G  H         H  J         J  VK         XK  L         L  M         M  
O         O  FP         HP  Q         Q  R         R  S         S  6U         8U  rV         tV  W         W  X         X  &Z         (Z  b[         d[  \         \  ]         ]  _         _  R`         T`  a         a  b         b  d         d  Be         De  ~f         f  g         g  h         h  2j         4j  nk         pk  l         l  m         m  "o          Nu  Cv          zv  ow          w  x          x  y          y  z                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               Y!         Z!  2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 e
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     --    -;-x    ;-/J-|    /J-}}-\    }}--    --P    -b-    b-c-	    c-1-D
    1--׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               D!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                  *	         +	  U
         V
                                   (         )  S         T  ~                                              0  1          k  l                                               !  !          ""  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            0         2  b         d                                   *         ,  \         ^  	         	  
         
             $         &  V         X                                               P         R                                              J         L  |         ~                                   D          F   v!         x!  "         "  #         #  %         %  >&         @&  p'         r'  (         (  )         )  +         +  8,         :,  j-         l-  .         .  /         /   1         1  22         42  d3         f3  4         4  5         5  6         6  ,8         .8  ^9         `9  :         :  ;         ;  <         <  &>         (>  X?         Z?  @         @  A         A  B         B   D         "D  RE         TE  F         F  G         G  H         H  J         ir  s         s  t         t  u         u  0w         1w  bx         cx  y         y  z         z  {         {  *}         +}  \~         ]~                                   $         %  V         W                                              P         Q                                              J                      ҙ  ۚ                      X  a                               >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               K!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 i
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -u,-X     u,-+-     +--    -76-x    76-D-|    D-x-\    x-1-    1--h    -ib-    ib-m-	    m--h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     T                       D!  1          1  j=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     :            W  8          t  U                                    	          H	  L
          
                                   H         I  u         v                         -         .  Z         [                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            4         6  j         l                                   B         D  x         z  	         	  
         
             P         R                                   (         *  ^         `                                    6         8  l         n                                   D         F  z         |               !         !  #         #  R$          0  0          21  2          X2  B3          ~3  h4          4  5          I  UJ         VJ  K         K  L         L  M         M  -O         .O  cP         dP  Q         Q  R         R  T         T  ;U         <U  qV         rV  W         W  X         X  Z         Z  I[         J[  \         \  ]         ]  ^         ^  !`         "`  Wa         Xa  b         b  c         c  d         d  /f         0f  eg         fg  h         h  i         i  k         k  =l         >l  sm         tm  n         n  o         o  q         q  Kr         Lr  s         s  t         t  u         u  #w         $w  Yx         Zx  y         y  z         z  {         {  1}         2}  g~         h~             Ӏ         Ԁ  	         
  ?         @  u         v                                   M         N                                   %         &  [         \             Ǒ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               C!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     +
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,-O-     O--    -);-x    );-I-|    I-}-\    }--    --h    -]b-    ]b-ǂ-	    ǂ--h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     e            !          R!  1          1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             ,         -  Y         Z                                                                                  6            S  M            z                                                                     K          L   x!         y!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            6         8  n         p                                   N         P             	         	  
         
  .         0  f         h                                   F         H  ~                                 &         (  ^         `                                   >         @  v         x                          "          "  V#         X#  $         $  %         %  &          '  7(         8(  o)         p)  *         *  +         +  -         -  O.         P.  /         /  0         0  1         1  /3         03  g4         h4  5         5  6         6  8         8  G9         H9  :         :  ;         ;  <         <  '>         (>  _?         `?  @         @  A         A  C         C  ?D         @D  wE         xE  F         F  G         G  I          Xa  Rb          b  c          c  d          d  e          0f  *g         o   q         q  8r         :r  ps         rs  t         t  u         u  w         w  Px         Rx  y         y  z         z  {         {  0}         2}  h~         j~             ؀         ڀ             H         J                                   (         *  `         b             Ћ         ҋ           
  @         B  x         z                                  "  X                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               =!         >!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -u,-X     u,-+-     +--    -76-x    76-D-|    D-x-\    x-1-    1--h    -ib-    ib-m-	    m--h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     U                       D!  1          1  k=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     :            W  9          t  V                                    	          H	  K
          
                                   H         I  u         v                         -         .  Z         [                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            5         6  k         l                                   C         D  y         z  	         	  
         
             Q         R                                   )         *  _         `                                   7         8  m         n                                   E         F  {         |               !         !  #         #  S$          5  6          6  7          8   9          <9  &:          b:  L;          I  TJ         VJ  K         K  L         L  M         M  ,O         .O  bP         dP  Q         Q  R         R  T         T  :U         <U  pV         rV  W         W  X         X  Z         Z  H[         J[  ~\         \  ]         ]  ^         ^   `         "`  Va         Xa  b         b  c         c  d         d  .f         0f  dg         fg  h         h  i         i  k         k  <l         >l  rm         tm  n         n  o         o  q         q  Jr         Lr  s         s  t         t  u         u  "w         $w  Xx         Zx  y         y  z         z  {         {  0}         2}  f~         h~             Ҁ         Ԁ           
  >         @  t         v                                   L         N                                   $         &  Z         \             Ƒ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               C!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     *
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,-O-     O--    -);-x    );-I-|    I-}-\    }--    --h    -]b-    ]b-ǂ-	    ǂ--h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     d            !          R!  1          1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             ,         -  Y         Z                                                                                  6            S  L            y                                                                      K          L   x!         y!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            7         8  o         p                                   O         P             	         	  
         
  /         0  g         h                                   G         H                                   '         (  _         `                                   ?         @  w         x                          "          "  W#         X#  $         $  %         %  &          '  6(         8(  n)         p)  *         *  +         +  -         -  N.         P.  /         /  0         0  1         1  .3         03  f4         h4  5         5  6         6  8         8  F9         H9  ~:         :  ;         ;  <         <  &>         (>  ^?         `?  @         @  A         A  C         C  >D         @D  vE         xE  F         F  G         G  I          fg  `h          h  i          i  j          k  l          >l  8m         o  q         q  9r         :r  qs         rs  t         t  u         u  w         w  Qx         Rx  y         y  z         z  {         {  1}         2}  i~         j~             ـ         ڀ             I         J                                   )         *  a         b             ы         ҋ  	         
  A         B  y         z                        !         "  Y                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               =!         >!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,-7-     7--    -:-x    :-I-|    I-|-\    |-
-    
--l    -ob-    ob-Sx-     Sx--  	   --  	                  =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     V                       H!  1          1  r=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      :         ;  g	         h	  
         
           S                                              3          4            Q  -          n  J            g                                                           L   <!          y!  i"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7      	   operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1                    >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               D!         F!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 c
             |            2         4  f         h                                   6         8  j         l  	         	  
         
             :         <  n         p                        
           >         @  r         t                                   B         D  v         x                                   F         H  z          |   !         !  "         "  $         $  J%         L%  ~&         &  '         '  (         (  *         *  N+         P+  ,         ,  -         -  .         .  0          0  R1         T1  2         2  3         3  4         4  "6         $6  V7         X7  8         8  9         9  :         :  &<         (<  Z=         \=  >         >  ?         ?  @         @  *B         ,B  ^C         `C  D         D  E         E  F         F  .H         ^H  I         I  J         J  K         K  -M         .M  aN         bN  O         O  P         P  Q         Q  1S         2S  eT         fT  U         U  V         V  X         X  5Y         6Y  iZ         jZ  [         [  \         \  ^         ^  9_         :_  m`         n`  a         a  b         b  	d         
d  =e         >e  qf         rf  g         g  h         h  j         j  Ak         Bk  ul                        ׇ                      N  C            y            
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                 
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            4     -,-`     ,-U-     U--    -;-    ;-/J-    /J-}}-d    }}-7-    7--0    --    -b-H	    b-c-P
    c--
    --                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     e            !          T!  1          1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 	         	  D
         E
  q                                 )         *  V         W              ~                                  ;  5          x  r                                               !  !          )"  
#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q   w  operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       \            9         :  s         t                        !         "  [         \             	         	  	         
  C         D  }         ~                        +         ,  e         f                                   M         N                                   5         6  o         p                        !         !  W"         X"  #         #  $         $  &         &  ?'         @'  y(         z(  )         )  *         *  ',         (,  a-         b-  .         .  /         /  1         1  I2         J2  3         3  4         4  5         5  17         27  k8         l8  9         9  :         :  <         <  S=         T=  >         >  ?         ?  A         A  ;B         <B  uC         vC  D         D  E         E  #G          $G  %H          vl  wm          m  n          n  o          $p  %q         ^q  r         r  s         s  
u         u  Dv         Fv  ~w         w  x         x  y         y  ,{         .{  f|         h|  }         }  ~         ~             N         P                      ă             6         8  p         r                                    X         Z             ̎                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               R!         \!   2         
2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 g
                     
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -{,-X     {,-1-     1--    -:-x    :-I-|    I-|-\    |--    --h    -ub-    ub-Sx-	    Sx--h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     U                       D!  1          1  k=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      :         ;  g	         h	  
         
           S                                              3          4            Q  .          n  K            h                                                           L   ;!          y!  h"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            3         4  g         h                                   7         8  k         l  	         	  
         
             ;         <  o         p                                   ?         @  s         t                                   C         D  w         x                                   G         H  {          |   !         !  "         "  $         $  K%         L%  &         &  '         '  (         (  *         *  O+         P+  ,         ,  -         -  .         .  0          0  S1         T1  2         2  3         3  4         4  #6         $6  W7         X7  8         8  9         9  :         :  '<         (<  [=         \=  >         >  ?         ?  @         @  +B         ,B  _C         `C  D         D  E         E  F         F  /H         ^H  I         I  J         J  K         K  ,M         .M  `N         bN  O         O  P         P  Q         Q  0S         2S  dT         fT  U         U  V         V   X         X  4Y         6Y  hZ         jZ  [         [  \         \  ^         ^  8_         :_  l`         n`  a         a  b         b  d         
d  <e         >e  pf         rf  g         g  h         h  j         j  @k         Bk  tl                                  &            \  Q                               >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               F!         J!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 c
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            4     -,-`     ,-U-     U--    -;-    ;-/J-    /J-}}-d    }}-=-    =--0    --    -b-H	    b-c-P
    c--
    --                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     d            !          R!  1          1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 	         	  D
         E
  q                                 )         *  V         W              }                                  ;  4          x  q                                               !  !          )"  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q   w  operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       \            8         :  r         t                                  "  Z         \             	         	           
  B         D  |         ~                        *         ,  d         f                                   L         N                                   4         6  n         p                        !         !  V"         X"  #         #  $         $  &         &  >'         @'  x(         z(  )         )  *         *  &,         (,  `-         b-  .         .  /         /  1         1  H2         J2  3         3  4         4  5         5  07         27  j8         l8  9         9  :         :  <         <  R=         T=  >         >  ?         ?   A         A  :B         <B  tC         vC  D         D  E         E  "G         $G  ]H         vl  m         m  n         n  #p         $p  ]q          ^q  _r          r  s          s  t          u  v          Fv  Gw         w  x         x  y         y  -{         .{  g|         h|  }         }  ~         ~             O         P             Ã         ă             7         8  q         r                                    Y         Z             ͎                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               P!         X!  1         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 g
                     
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,-s-     s--    -C6-x    C6-D-|    D-+x-\    +x--    -w-h    w-b-    b-Yx-	    Yx--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               3!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     :            W  8          t  U                                    	          H	  L
          
                                   H         I  u         v                         -         .  Z         [                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       C  I                                             $  %         %  '         '  8(         :(  j)         l)  *         *  +         +   -         -  2.         4.  d/         f/  0         0  1         1  2         2  ,4         .4  ^5         `5  6         6  7         7  8         8  &:         (:  X;         Z;  <         <  =         =  >         >   @         "@  RA         TA  B         B  C         C  D         D  F         F  LG         NG  ~H         H  I         I  K         K  BL         DL  tM         vM  N         N  O         O  
Q         Q  <R         >R  nS         pS  T         T  U         U  W         W  7X         8X  iY         jY  Z         Z  [         [  \          ]  1^         2^  c_         d_  `         `  a         a  b         b  +d         ,d  ]e         ^e  f         f  g         g  h         h  %j         &j  Wk         Xk  l         l  m         m  n         n  p          p  Qq         Rq  r         r  s         s  t         t  v         v  Kw         Lw  }x         ~x  y         y  z         z  |         |  E}         F}  w~         x~             ۀ         ܀             ?         @  q         r             Ն         ֆ             9         :  k         l             ό         Ќ             3         d                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     T                       8!  1          1  X=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 c
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -#-    #-5;-    5;-I-    I-}-d    }--    --p    -ib-    ib---
    --;-
    ;--t    --׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               A!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             ,         -  Y         Z                                                                                  6            S  M            z                                                                     K          L   x!         y!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                             "            L  =         T#  $         $  %         %  '         '  C(         D(  )         )  *         *  +         +  3-         4-  o.         p.  /         /  0         0  #2         $2  _3         `3  4         4  5         5  7         7  O8         P8  9         9  :         :  <         <  ?=         @=  {>         |>  ?         ?  @         @  /B         0B  kC         lC  D         D  E         E  G          G  [H         \H  I         J  K         K  8M         :M  tN         vN  O         O  P         P  (R         *R  dS         fS  T         T  U         U  W         W  TX         VX  Y         Y  Z         Z  \         
\  D]         F]  ^         ^  _         _  `         `  4b         6b  pc         rc  d         d  e         e  $g         &g  `h         bh  i         i  j         j  l         l  Pm         Rm  n         n  o         o  q         q  @r         Br  |s         ~s  t         t  u         u  0w         2w  lx         nx  y         y  z         z   |         "|  \}         ^}  ~         ~                        L         N             Ą         Ƅ              <         >  x         z                        ,         .  h         j                                   X         Z                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     e            !          f!  1          2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    t
         u
             ^      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                      
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-m-     m--    -C6-x    C6-D-|    D-+x-\    +x--    -}-h    }-b-    b-Yx-	    Yx--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               1!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     :            W  9          t  V                                    	          H	  K
          
                                   H         I  u         v                         -         .  Z         [                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         O  U                        	          
            [  a         $  %         %  '         '  9(         :(  k)         l)  *         *  +         +  -         -  3.         4.  e/         f/  0         0  1         1  2         2  -4         .4  _5         `5  6         6  7         7  8         8  ':         (:  Y;         Z;  <         <  =         =  >         >  !@         "@  SA         TA  B         B  C         C  D         D  F         F  MG         NG  H         H  I         I  K         K  CL         DL  uM         vM  N         N  O         O  Q         Q  =R         >R  oS         pS  T         T  U         U  W         W  6X         8X  hY         jY  Z         Z  [         [  \          ]  0^         2^  b_         d_  `         `  a         a  b         b  *d         ,d  \e         ^e  f         f  g         g  h         h  $j         &j  Vk         Xk  l         l  m         m  n         n  p          p  Pq         Rq  r         r  s         s  t         t  v         v  Jw         Lw  |x         ~x  y         y  z         z  |         |  D}         F}  v~         x~             ڀ         ܀             >         @  p         r             Ԇ         ֆ             8         :  j         l             Ό         Ќ              2         d                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     U                       <!  1          1  _=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 c
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -#-    #-5;-    5;-I-    I-}-d    }--    --p    -ib-    ib---
    --;-
    ;--t    --׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ?!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             ,         -  Y         Z                                                                                  6            S  L            y                                                                      K          L   x!         y!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         v  g                                                         T#  $         $  %         %  '         '  B(         D(  ~)         )  *         *  +         +  2-         4-  n.         p.  /         /  0         0  "2         $2  ^3         `3  4         4  5         5  7         7  N8         P8  9         9  :         :  <         <  >=         @=  z>         |>  ?         ?  @         @  .B         0B  jC         lC  D         D  E         E  G          G  ZH         \H  I         J  K         K  9M         :M  uN         vN  O         O  P         P  )R         *R  eS         fS  T         T  U         U  W         W  UX         VX  Y         Y  Z         Z  	\         
\  E]         F]  ^         ^  _         _  `         `  5b         6b  qc         rc  d         d  e         e  %g         &g  ah         bh  i         i  j         j  l         l  Qm         Rm  n         n  o         o  q         q  Ar         Br  }s         ~s  t         t  u         u  1w         2w  mx         nx  y         y  z         z  !|         "|  ]}         ^}  ~         ~                        M         N             ń         Ƅ             =         >  y         z                        -         .  i         j                                   Y         Z                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     d            !          f!  1          2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    t
         u
             ^      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                      
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-O-     O--    -:-x    :-I-|    I-|-\    |-1-    1--h    -'b-    'b--	    --h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               "!         4!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      :         ;  g	         h	  
         
           S                                              3          4            Q  -          n  J            g                                                           L   <!          y!  i"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            .         0  ^         `                                               N         P  ~	         	  
         
                        >         @  n         p                                    .         0  ^         `                                               N         P  ~                                              >!         @!  n"         p"  #         #  $         $  %          &  .'         0'  ^(         `(  )         )  *         *  +         +  -          -  N.         P.  ~/         /  0         0  1         1  3         3  >4         @4  n5         p5  6         6  7         7  8          9  .:         0:  ^;         `;  <         <  =         =  >         >  @          @  NA         PA  ~B         B  C         C  D         D  F         F  >G         @G  nH         pH  I         I  J          tV  `W          W  X          X  Y          Y  Z          [  \         d                      Ĕ             "         $  R         T                                              B         D  r         t             ҡ         ԡ             2         4  b         d             §         ħ             "         $  R         T                                              B         D  r                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     Q                       D!  1          1  g=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   "      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     -/-    /-S5-    S5-C-    C-;w-l    ;w-A4-    A4--x    -b-	    b-3-
    3-A-
    A--|    -ճ-    ճ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               C!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         -  '          Z  T                                                            	          8	  
          U
  ;         r                                   %         &  R                      =         >  j         k                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |        (#  c$          :  ;          <  =          ;=  #>          \>  D?          }?  e@         CF  ~G         G  H         H  I         I  8K         :K  vL         xL  M         M  N         N  0P         2P  nQ         pQ  R         R  S         S  (U         *U  fV         hV  W         W  X         X   Z         "Z  ^[         `[  \         \  ]         ]  _         _  V`         X`  a         a  b         b  d         d  Ne         Pe  f         f  g         g  i         
i  Fj         Hj  k         k  l         l   n         n  >o         @o  |p         ~p  q         q  r         r  6t         8t  tu         vu  v         v  w         w  .y         0y  lz         nz  {         {  |         |  &~         (~  d         f                                    \         ^             ؆         چ             T         V             Ћ         ҋ             L         N             Ȑ         ʐ             C         E                                   9         ;  w         y                        1         3  o         q                        )         +  g         i                        !         #  _         a             ۪         ݪ             W         Y             ӯ         կ             O         Q             ˴                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               D!         T!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    t
         u
             ^      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    t
         u
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            0     -,-X     ,-O-     O--    -:-x    :-I-|    I-|-\    |-1-    1--h    -!b-    !b--	    --h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               !!         2!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      :         ;  g	         h	  
         
           S                                              3          4            Q  .          n  K            h                                                           L   ;!          y!  h"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            /         0  _         `                                               O         P  	         	  
         
                        ?         @  o         p                                    /         0  _         `                                               O         P                                                ?!         @!  o"         p"  #         #  $         $  %          &  /'         0'  _(         `(  )         )  *         *  +         +  -          -  O.         P.  /         /  0         0  1         1  3         3  ?4         @4  o5         p5  6         6  7         7  8          9  /:         0:  _;         `;  <         <  =         =  >         >  @          @  OA         PA  B         B  C         C  D         D  F         F  ?G         @G  oH         pH  I         I  J          F\  2]          p]  \^          ^  _          _  `          `  a         d             Ô         Ĕ             #         $  S         T                                              C         D  s         t             ӡ         ԡ             3         4  c         d             ç         ħ             #         $  S         T                                              C         D  s                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     P                       @!  1          1  `=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   !      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     -/-    /-S5-    S5-C-    C-;w-l    ;w-A4-    A4--x    -b-	    b-3-
    3-A-
    A--|    -ճ-    ճ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               A!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         -  &          Z  S                                                            	          8	  
          U
  <         r                                   %         &  R                      =         >  j         k                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |        (#  d$          @  A          A  B          B  C          D  D          "E  
F         CF  G         G  H         H  I         I  9K         :K  wL         xL  M         M  N         N  1P         2P  oQ         pQ  R         R  S         S  )U         *U  gV         hV  W         W  X         X  !Z         "Z  _[         `[  \         \  ]         ]  _         _  W`         X`  a         a  b         b  d         d  Oe         Pe  f         f  g         g  	i         
i  Gj         Hj  k         k  l         l  n         n  ?o         @o  }p         ~p  q         q  r         r  7t         8t  uu         vu  v         v  w         w  /y         0y  mz         nz  {         {  |         |  '~         (~  e         f                                    ]         ^             ن         چ             U         V             ы         ҋ             M         N             ɐ         ʐ             D         E                                   :         ;  x         y                        2         3  p         q                        *         +  h         i                        "         #  `         a             ܪ         ݪ             X         Y             ԯ         կ             P         Q             ̴                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               @!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    t
         u
             ^      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    t
         u
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            0     -,-X     ,--     -#-    #-;;-x    ;;-I-|    I-#}-\    #}--    --h    -ib-    ib--	    --h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ;!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             ,         -  Y         Z                                  r  b                                              &            S  4          p  Q            n                                                         K          L   x!         y!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ,         .  Z         \                                              @         B  n	         p	  
         
                        &         (  T         V                                              :         <  h         j                                             "  N         P  |         ~                        !         !  4"         6"  b#         d#  $         $  %         %  &         &  (         (  H)         J)  v*         x*  +         +  ,         ,   .         .  ./         0/  \0         ^0  1         1  2         2  3         3  5         5  B6         D6  p7         r7  8         8  9         9  :         :  (<         *<  V=         X=  >         >  ?         ?  @         @  B         B  <C         >C  jD         lD  E         E  F          \]  ?^          }^  `_          _  `          `  a          a  b         l  m         m  o          o  Mp         Np  {q         |q  r         r  s         s  u         u  3v         4v  aw         bw  x         x  y         y  z         z  |         |  G}         H}  u~         v~             р         Ҁ              -         .  [         \                                              A         B  o         p             ˍ         ̍             '                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               /!         J!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   "      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -G-    G-;-x    ;-;J-|    ;J-}-\    }-m-    m--h    -b-    b-x-	    x--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               L!         \!  1         
2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 	         	  D
         E
  q                                 )         *  V         W              ~                                  ;  5          x  r                                               !  !          )"  
#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            -         .  [         \                                              A         B  o	         p	  
         
                        '         (  U         V                                              ;         <  i         j                                   !         "  O         P  }         ~                        !         !  5"         6"  c#         d#  $         $  %         %  &         &  (         (  H)         J)  v*         x*  +         +  ,         ,   .         .  ./         0/  \0         ^0  1         1  2         2  3         3  5         5  B6         D6  p7         r7  8         8  9         9  :         :  (<         *<  V=         X=  >         >  ?         ?  @         @  B         B  <C         >C  jD         lD  E         E  F         F  G         G  "I         $I  PJ         RJ  ~K         K  L         L  M         M  O         
O  6P         8P  dQ         fQ  R         R  S         S  T         T  V         V  JW         LW  xX         zX  Y         Y  Z         Z  \         \  0]         2]  ^^         `^  _         _  `         `  a         a  c         c  Dd         Fd  re         te  f         f  g         g  h          l  m          n  n          @o  :p          ~p  xq          q  r         ʐ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               G!         Z!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 w
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-;;-x    ;;-I-|    I-#}-\    #}--    --h    -ib-    ib--	    --h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ?!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             ,         -  Y         Z                                  r  a                                              &            S  5          p  R            o                                                         K          L   x!         y!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            -         .  [         \                                              A         B  o	         p	  
         
                        '         (  U         V                                              ;         <  i         j                                   !         "  O         P  }         ~                        !         !  5"         6"  c#         d#  $         $  %         %  &         &  (         (  I)         J)  w*         x*  +         +  ,         ,  .         .  //         0/  ]0         ^0  1         1  2         2  3         3  5         5  C6         D6  q7         r7  8         8  9         9  :         :  )<         *<  W=         X=  >         >  ?         ?  @         @  B         B  =C         >C  kD         lD  E         E  F          c  c          "d  e          Ce  &f          df  Gg          g  hh         l  m         m  o          o  Lp         Np  zq         |q  r         r  s         s  u         u  2v         4v  `w         bw  x         x  y         y  z         z  |         |  F}         H}  t~         v~             Ѐ         Ҁ              ,         .  Z         \                                              @         B  n         p             ʍ         ̍             &                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               /!         J!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   !      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -;-    ;-;-x    ;-;J-|    ;J-}-\    }-s-    s--h    -b-    b-x-	    x--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               H!         X!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 	         	  D
         E
  q                                 )         *  V         W              }                                  ;  4          x  q                                               !  !          )"  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ,         .  Z         \                                              @         B  n	         p	  
         
                        &         (  T         V                                              :         <  h         j                                             "  N         P  |         ~                        !         !  4"         6"  b#         d#  $         $  %         %  &         &  (         (  I)         J)  w*         x*  +         +  ,         ,  .         .  //         0/  ]0         ^0  1         1  2         2  3         3  5         5  C6         D6  q7         r7  8         8  9         9  :         :  )<         *<  W=         X=  >         >  ?         ?  @         @  B         B  =C         >C  kD         lD  E         E  F         F  G         G  #I         $I  QJ         RJ  K         K  L         L  M         M  	O         
O  7P         8P  eQ         fQ  R         R  S         S  T         T  V         V  KW         LW  yX         zX  Y         Y  Z         Z  \         \  1]         2]  _^         `^  _         _  `         `  a         a  c         c  Ed         Fd  se         te  f         f  g         g  h          r  s          8t  2u          vu  pv          v  w          w  x         ʐ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               E!         Z!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 w
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    <     -,-p     ,--     -/-    /-O6-    O6-D-    D-7x-t    7x-8-     8-w@-X    w@--    -b-p	    b-9-x
    9-G-
    G--    -ճ-\    ճ--<                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               =!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     :            W  8          t  U                                    	          H	  L
          
                                   H         I  u         v                         -         .  Z         [                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h         zI  J          J  K          K  L          4M  9N          rN  wO         O  P         P  *R         ,R  hS         jS  T         T  U         U  "W         $W  `X         bX  Y         Y  Z         Z  \         \  X]         Z]  ^         ^  _         _  a         a  Pb         Rb  c         c  d         d  
f         f  Hg         Jg  h         h  i         i  k         k  Al         Bl  m         m  n         n  o         o  9q         :q  wr         xr  s         s  t         t  1v         2v  ow         pw  x         x  y         y  ){         *{  g|         h|  }         }  ~         ~  !         "  _         `             ۃ         ܃             W         X             ӈ         Ԉ             O         P             ˍ         ̍  	         
  G         H             Ò         Ē             ?         @  }         ~                        7         8  u         v                        /         0  m         n                        '         (  e         f                                    ]         ^             ٬         ڬ             U         V             ѱ         ұ             L         N             ȶ         ޶             Y                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               >!         P!  1         1  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    t
         u
             ^                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    t
         u
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            0     -,-X     ,--     -M-    M-_5-x    _5-C-|    C-Gw-\    Gw-i0-    i0--h    -b-    b-x-	    x--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               N!         \!  1         
2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         -  '          Z  T                                                            	          8	  
          U
  ;         r                                   %         &  R                      =         >  j         k                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                 $  0          g  s                     %  	'         
'  9(         :(  i)         j)  *         *  +         +  ,         ,  ).         *.  Y/         Z/  0         0  1         1  2         2  4         4  I5         J5  y6         z6  7         7  8         8  	:         
:  9;         :;  i<         j<  =         =  >         >  ?         ?  )A         *A  YB         ZB  C         C  D         D  E         E  G         G  IH         JH  yI         h  i         i  *k         ,k  Zl         \l  m         m  n         n  o         o  q         q  Jr         Lr  zs         |s  t         t  u         u  
w         w  :x         <x  jy         ly  z         z  {         {  |         |  *~         ,~  Z         \                                              J         L  z         |             ڈ         ܈  
           :         <  j         P  ~                      ސ                      >         @  n         p             Ζ         Ж              .         0  ^         `                                               N         P  ~                      ޣ                      >         @  n         p             Ω         Щ              .         0  ^         `                                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               B!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 w
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    <     -,-p     ,--     -5-    5-O6-    O6-D-    D-7x-t    7x-8-     8-q@-X    q@--    -b-p	    b-9-x
    9-G-
    G--    -ճ-\    ճ--<                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               A!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     :            W  9          t  V                                    	          H	  K
          
                                   H         I  u         v                         -         .  Z         [                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        zI  J         J  K         K  3M         4M  qN         rN  O          O  P          P  Q          ,R  1S          jS  oT          T  U         U  #W         $W  aX         bX  Y         Y  Z         Z  \         \  Y]         Z]  ^         ^  _         _  a         a  Qb         Rb  c         c  d         d  f         f  Ig         Jg  h         h  i         i  k         k  @l         Bl  ~m         m  n         n  o         o  8q         :q  vr         xr  s         s  t         t  0v         2v  nw         pw  x         x  y         y  ({         *{  f|         h|  }         }  ~         ~            "  ^         `             ڃ         ܃             V         X             ҈         Ԉ             N         P             ʍ         ̍           
  F         H                      Ē              >         @  |         ~                        6         8  t         v                        .         0  l         n                        &         (  d         f                                    \         ^             ج         ڬ             T         V             б         ұ             M         N             ɶ         ޶             X                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               @!         P!  1         1  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    t
         u
             ^                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    t
         u
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            0     -,-X     ,--     -A-    A-_5-x    _5-C-|    C-Gw-\    Gw-o0-    o0--h    -b-    b-x-	    x--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               J!         X!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         -  &          Z  S                                                            	          8	  
          U
  <         r                                   %         &  R                      =         >  j         k                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                     0  <          s                                   %  '         
'  8(         :(  h)         j)  *         *  +         +  ,         ,  (.         *.  X/         Z/  0         0  1         1  2         2  4         4  H5         J5  x6         z6  7         7  8         8  :         
:  8;         :;  h<         j<  =         =  >         >  ?         ?  (A         *A  XB         ZB  C         C  D         D  E         E  G         G  HH         JH  xI         h  i         i  +k         ,k  [l         \l  m         m  n         n  o         o  q         q  Kr         Lr  {s         |s  t         t  u         u  w         w  ;x         <x  ky         ly  z         z  {         {  |         |  +~         ,~  [         \                                              K         L  {         |             ۈ         ܈             ;         <  k         P                        ߐ                      ?         @  o         p             ϖ         Ж              /         0  _         `                                               O         P                        ߣ                      ?         @  o         p             ϩ         Щ              /         0  _         `                                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               B!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 w
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-A;-x    A;-I-|    I-)}-\    )}-!9-    !9--h    -b-    b-	n-	    	n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               >!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             ,         -  Y         Z                                  r  b                                              &            S  4          p  Q            n                                                         K          L   x!         y!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            :         <  v         x                        *         ,  f         h             	         	             V         X                        
           F         H                                   6         8  r         t                        &         (  b         d                                     R!         T!  "         "  #         #  %         %  C&         D&  '         '  (         (  )         )  3+         4+  o,         p,  -         -  .         .  #0         $0  _1         `1  2         2  3         3  5         5  O6         P6  7         7  8         8  :         :  ?;         @;  {<         |<  =         =  >         >  /@         0@  kA         lA  B         B  C         C  E          E  [F         \F  G          T  U          U  V          W  X          RX  YY          Y  Z         m  o         o  Op           ֕         ؕ             N         P             ƚ         Ț             >         @  z         |                        .         0  j         l                                    Z         \             Ҫ         Ԫ             J         L             ¯         į              :         <  v         x                        *         ,  f         h                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               A!         V!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     +
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -A-    A-;-x    ;-AJ-|    AJ-}-\    }-I-    I--h    -ob-    ob-߂-	    ߂--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               F!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 	         	  D
         E
  q                                 )         *  V         W              ~                                  ;  5          x  r                                               !  !          )"  
#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            1         2  c         d                                   +         ,  ]         ^  	         	  
         
             %         &  W         X                                               Q         R                                              K         L  }         ~                                   E          F   w!         x!  "         "  #          \:  8;          |;  X<          <  x=          =  >          >  ?         G  H         H  I         I  -K         .K  _L         `L  M         M  N         N  O         O  'Q         (Q  YR         ZR  S         S  T         T  U         U  !W         "W  SX         TX  Y         Y  Z         Z  [         [  ]         ]  M^         N^  _         _  `         `  a         a  c         c  Gd         Hd  ye         ze  f         f  g         g  i         i  Aj         Bj  sk         tk  l         l  m         Pp  q         q  r         r  s         s  u         u  Hv         Jv  zw         |w  x         x  y         y  {         {  B|         D|  t}         v}  ~         ~             
           <         >  n         p             ҅         ԅ             6         8  h         j             ̋         ΋              0         2  b         d             Ƒ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               1!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-A;-x    A;-I-|    I-)}-\    )}-'9-    '9--h    -b-    b-	n-	    	n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               @!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             ,         -  Y         Z                                  r  a                                              &            S  5          p  R            o                                                         K          L   x!         y!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ;         <  w         x                        +         ,  g         h             	         	             W         X                                   G         H                                   7         8  s         t                        '         (  c         d                                     S!         T!  "         "  #         #  %         %  B&         D&  ~'         '  (         (  )         )  2+         4+  n,         p,  -         -  .         .  "0         $0  ^1         `1  2         2  3         3  5         5  N6         P6  7         7  8         8  :         :  >;         @;  z<         |<  =         =  >         >  .@         0@  jA         lA  B         B  C         C  E          E  ZF         \F  G          Z  [          \  "]          ^]  e^          ^  _          _  `         m  o         o  Np           ו         ؕ             O         P             ǚ         Ț             ?         @  {         |                        /         0  k         l                                    [         \             Ӫ         Ԫ             K         L             ï         į              ;         <  w         x                        +         ,  g         h                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               E!         Z!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     *
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -A-    A-;-x    ;-AJ-|    AJ-}-\    }-I-    I--h    -ob-    ob-߂-	    ߂--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               F!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 	         	  D
         E
  q                                 )         *  V         W              }                                  ;  4          x  q                                               !  !          )"  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            0         2  b         d                                   *         ,  \         ^  	         	  
         
             $         &  V         X                                               P         R                                              J         L  |         ~                                   D          F   v!         x!  "         "  #          ?  @          A  A          <B  C          \C  8D          |D  XE         G  H         H  I         I  ,K         .K  ^L         `L  M         M  N         N  O         O  &Q         (Q  XR         ZR  S         S  T         T  U         U   W         "W  RX         TX  Y         Y  Z         Z  [         [  ]         ]  L^         N^  ~_         _  `         `  a         a  c         c  Fd         Hd  xe         ze  f         f  g         g  i         i  @j         Bj  rk         tk  l         l  m         Pp  q         q  r         r  s         s  u         u  Iv         Jv  {w         |w  x         x  y         y  {         {  C|         D|  u}         v}  ~         ~                        =         >  o         p             Ӆ         ԅ             7         8  i         j             ͋         ΋              1         2  c         d             Ǒ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               1!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-y6-x    y6-E-|    E-ax-\    ax--    --h    -b-    b-n-	    n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               <!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     :            W  8          t  U                                    	          H	  I
          
                                   H         I  u         v                        ;         <  o         p                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            8         :  r         t                                  "  Z         \             	         	           
  B         D  |         ~                        *         ,  d         f                                   L         N                                   4         6  n         p                        !         !  V"         X"  #         #  $          <  =          >  ?          D?  <@          x@  pA          A  B         H  J         J  WK         XK  L         L  M         M  O         O  ?P         @P  yQ         zQ  R         R  S         S  'U         (U  aV         bV  W         W  X         X  Z         Z  I[         J[  \         \  ]         ]  ^         ^  1`         2`  ka         la  b         b  c         c  e         e  Sf         Tf  g         g  h         h  j         j  ;k         <k  ul         vl  m         No  p         p  q         q  r         r  5t         6t  ou         pu  v         v  w         w  y         y  Wz         Xz  {         {  |         |  ~         ~  ?         @  y         z                        '         (  a         b             Շ         և             I         J                                   1         2  k         l             ߒ                              >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               <!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     +
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -A-    A-k;-x    k;-J-|    J-S}-\    S}--    --h    -]b-    ]b-߂-	    ߂--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               F!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             ,         -  Y         Z                                                                                  6            S  H            u                                                          A         B  n          o   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            2         4  f         h                                   6         8  j         l  	         	  
         
             :         <  n         p                        
           >         @  r         t                                   B         D  v         x                                   F         H  z          |   !         !  "         "  $         $  %          &  3'         4'  g(         h(  )         )  *         *  ,         ,  7-         8-  k.         l.  /         /  0         0  2         2  ;3         <3  o4         p4  5         5  6         6  8         8  ?9         @9  s:         t:  ;         ;  <         <  >         >  C?         D?  w@         x@  A         A  B         B  D         D  GE         HE  {F         |F  G         G  H          H  I          J  K          XK  VL          L  M          M  N         "o  Tp         Vp  q         q  r         r  s         s  $u         &u  Xv         Zv  w         w  x         x  y         y  ({         *{  \|         ^|  }         }  ~         ~             ,         .  `         b             Ȅ         ʄ             0         2  d         f             ̊         Ί              4         6  h         j             А         Ґ             8                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               *!         2!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-s6-x    s6-E-|    E-[x-\    [x--    --h    -b-    b-n-	    n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               >!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     :            W  9          t  V                                    	          H	  J
          
                                   H         I  u         v                        9         :  l         m                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            9         :  s         t                        !         "  [         \             	         	  	         
  C         D  }         ~                        +         ,  e         f                                   M         N                                   5         6  o         p                        !         !  W"         X"  #         #  $          B  C          D  E          HE  @F          |F  tG          G  H         H  J         J  VK         XK  L         L  M         M  O         O  >P         @P  xQ         zQ  R         R  S         S  &U         (U  `V         bV  W         W  X         X  Z         Z  H[         J[  \         \  ]         ]  ^         ^  0`         2`  ja         la  b         b  c         c  e         e  Rf         Tf  g         g  h         h   j         j  :k         <k  tl         vl  m         No  p         p  q         q  r         r  4t         6t  nu         pu  v         v  w         w  y         y  Vz         Xz  {         {  |         |  ~         ~  >         @  x         z                        &         (  `         b             ԇ         և             H         J                                   0         2  j         l             ޒ                              >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               >!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     *
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -;-    ;-e;-x    e;-I-|    I-M}-\    M}--    --h    -cb-    cb-߂-	    ߂--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               B!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             ,         -  Y         Z                                                                                  6            S  G            t                                                          <         =  i          j   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            3         4  g         h                                   7         8  k         l  	         	  
         
             ;         <  o         p                                   ?         @  s         t                                   C         D  w         x                                   G         H  {          |   !         !  "         "  $         $  %          &  2'         4'  f(         h(  )         )  *         *  ,         ,  6-         8-  j.         l.  /         /  0         0  2         2  :3         <3  n4         p4  5         5  6         6  
8         8  >9         @9  r:         t:  ;         ;  <         <  >         >  B?         D?  v@         x@  A         A  B         B  D         D  FE         HE  zF         |F  G         G  H          O  P          @P  >Q          zQ  xR          R  S          S  T         "o  Up         Vp  q         q  r         r  s         s  %u         &u  Yv         Zv  w         w  x         x  y         y  ){         *{  ]|         ^|  }         }  ~         ~             -         .  a         b             Ʉ         ʄ             1         2  e         f             ͊         Ί             5         6  i         j             ѐ         Ґ             9                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ,!         6!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -S-    S-);-x    );-I-|    I-}-\    }--    --h    -b-    b-qx-	    qx--h
    -K׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               O!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      :         ;  g	         h	  
         
           S                                              3          4            X  0          |  T            x                                              B  '           o   T!          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            7         8  o         p                                   O         P             	         	  
         
  /         0  g         h                                   G         H                                   '         (  _         `                                   ?         @  w         x                          "          "  W#         X#  $         $  %         %  &          '  6(         8(  n)         p)  *         *  +         +  -         -  N.         P.  /         /  0         0  1         1  .3         03  f4         h4  5         5  6         6  8         8  F9         H9  ~:         :  ;         ;  <         <  &>         (>  ^?         `?  @         @  A         A  C         C  >D         @D  vE         xE  F         F  G         G  I         I  K         K  ML         NL  M         M  N         N  O         O  -Q         .Q  eR         fR  S         S  T         T  V         V  EW         FW  }X         ~X  Y         Y  Z         Z  %\         &\  ]]         ^]  ^         ^  _         _  a         a  =b         >b  uc         vc  d         d  e         e  g         g  Uh         Vh  i         i  j         j  k         k  5m         6m  mn          No  <p          p  vq          q  r          r  s          6t  $u                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               H!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 c
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -G-    G-;-x    ;-_J-|    _J-}-\    }-w%-    w%--P    -b-    b-c-	    c-=-D
    =-K׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               J!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 &	         '	  S
         T
                                   8         9  e         f                                                J  ;            x                                               !  !          8"  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            5         6  k         l                                   C         D  y         z  	         	  
         
             Q         R                                   )         *  _         `                                   7         8  m         n                                   E         F  {         |               !         !  #         #  S$         T$  %         %  &         &  '         '  +)         ,)  a*         b*  +         +  ,         ,  .         .  9/         :/  o0         p0  1         1  2         2  4         4  G5         H5  }6         ~6  7         7  8         8  :          :  U;         V;  <         <  =         =  >         >  -@         .@  cA         dA  B         B  C         C  E         E  ;F         <F  pG         rG  H         H  I         nn  o         o  p         p  r         r  Ds         Fs  zt         |t  u         u  v         v  x         x  Ry         Ty  z         z  {         {  |         |  *~         ,~  `         b             ́         ΁             8         :  n         p             ڇ         ܇             F         H  |         ~                       ؟  ߠ            "          ^  e                                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               B!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 g
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -G-    G-#;-x    #;-I-|    I-}-\    }--    --h    -b-    b-qx-	    qx--h
    -K׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               K!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      :         ;  g	         h	  
         
           S                                              3          4            W  /          z  R            u                                              =  !           j   N!          !  {"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            6         8  n         p                                   N         P             	         	  
         
  .         0  f         h                                   F         H  ~                                 &         (  ^         `                                   >         @  v         x                          "          "  V#         X#  $         $  %         %  &          '  7(         8(  o)         p)  *         *  +         +  -         -  O.         P.  /         /  0         0  1         1  /3         03  g4         h4  5         5  6         6  8         8  G9         H9  :         :  ;         ;  <         <  '>         (>  _?         `?  @         @  A         A  C         C  ?D         @D  wE         xE  F         F  G         G  I         I  K         K  LL         NL  M         M  N         N  O         O  ,Q         .Q  dR         fR  S         S  T         T  V         V  DW         FW  |X         ~X  Y         Y  Z         Z  $\         &\  \]         ^]  ^         ^  _         _  a         a  <b         >b  tc         vc  d         d  e         e  g         g  Th         Vh  i         i  j         j  k         k  4m         6m  ln          pu  ^v          v  w          w  x          y  z          Xz  F{                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               J!         R!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 c
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -G-    G-;-x    ;-eJ-|    eJ-}-\    }-+-    +--P    -b-    b-c-	    c-=-D
    =-K׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               I!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 +	         ,	  X
         Y
                                   =         >  j         k                                                O  A            ~                                                 !  "          ="  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            4         6  j         l                                   B         D  x         z  	         	  
         
             P         R                                   (         *  ^         `                                    6         8  l         n                                   D         F  z         |               !         !  #         #  R$         T$  %         %  &         &  '         '  *)         ,)  `*         b*  +         +  ,         ,  .         .  8/         :/  n0         p0  1         1  2         2  4         4  F5         H5  |6         ~6  7         7  8         8  :          :  T;         V;  <         <  =         =  >         >  ,@         .@  bA         dA  B         B  C         C  E         E  :F         <F  qG         rG  H         H  I         nn  o         o  p         p  r         r  Es         Fs  {t         |t  u         u  v         v  x         x  Sy         Ty  z         z  {         {  |         |  +~         ,~  a         b             ́         ΁             9         :  o         p             ۇ         ܇             G         H  }         ~                       '  .          j  q                                  3  :                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               D!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 g
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -M-    M-[6-x    [6-D-|    D-Cx-\    Cx-7-    7--h    -b-    b-wx-	    wx--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               O!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     :            W  8          t  U                                    	          H	  D
          
                                   H         I  u         v                         -         .  Z         [                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                     <  .          p  b                                 U$  %         %  &         &  '         '  ,)         -)  b*         c*  +         +  ,         ,  .         .  :/         ;/  p0         q0  1         1  2         2  4         4  H5         I5  ~6         6  7         7  8         8   :         !:  V;         W;  <         <  =         =  >         >  .@         /@  dA         eA  B         B  C         C  E         E  <F         =F  rG         G  I         I  CJ         EJ  yK         {K  L         L  M         M  O         O  QP         SP  Q         Q  R         R  S         S  )U         +U  _V         aV  W         W  X         X  Z         Z  7[         9[  m\         o\  ]         ]  ^         ^  `         `  Ea         Ga  {b         }b  c         c  d         d  f         f  Sg         Ug  h         h  i         i  j         j  +l         l  
n         ړ             E         F  {         |                                   S         T                                   +         ,  a         b             ͤ         Τ             9         :  o         p             ۪         ܪ             G         H  }         ~                                    U         V                      µ             -                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               C!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 c
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     -;-    ;-M;-    M;-I-    I-5}-l    5}--    -'-P    '--    -b-h	    b-E-p
    E-S-
    S-۰-    ۰-E׳-L                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               E!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             ,         -  Y         Z                                                                                  6            S  E            r                                                                     K          L   x!         y!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        T#  $         $  %         %  &         &  3(         4(  k)         l)  *         *  +         +  -         -  K.         L.  /         /  0         0  1         1  +3         ,3  c4         d4  5         5  6         6  8         8  C9         D9  {:         |:  ;         ;  <         <  #>         $>  [?         \?  @         @  A         A  C         C  ;D         <D  sE         tE  F         F  G         G  I         I  RJ         TJ  K         K  L         L  M         M  2O         4O  jP         lP  Q         Q  R         R  T         T  JU         LU  V         V  W         W  X         X  *Z         ,Z  b[         d[  \         \  ]         ]  
_         _  B`          D`  Ca          |a  {b          b  c          c  d          $e  #f         \f  g         g  h         h  j         j  :k         <k  rl         tl  m         m  n         n  p         p  Rq         Tq  r         r  s         s  t         t  2v         4v  jw         lw  x         x  y         y  {         {  J|         L|  }         }  ~         ~             *         ,  b         d             ҄         Ԅ  
           B         D  z         |                        "         $  Z         \             ʏ         ̏                                b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               M!         ^!  1         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    t
         u
             ^                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     &
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -G-    G-[6-x    [6-D-|    D-Cx-\    Cx-7-    7--h    -b-    b-wx-	    wx--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               K!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     :            W  9          t  V                                    	          H	  C
          
                                   H         I  u         v                         -         .  Z         [                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                     @  2          t  f                                 U$  %         %  &         &  '         '  +)         -)  a*         c*  +         +  ,         ,  .         .  9/         ;/  o0         q0  1         1  2         2  4         4  G5         I5  }6         6  7         7  8         8  :         !:  U;         W;  <         <  =         =  >         >  -@         /@  cA         eA  B         B  C         C  E         E  ;F         =F  qG         G  I         I  DJ         EJ  zK         {K  L         L  M         M  O         O  RP         SP  Q         Q  R         R  S         S  *U         +U  `V         aV  W         W  X         X  Z         Z  8[         9[  n\         o\  ]         ]  ^         ^  `         `  Fa         Ga  |b         }b  c         c  d         d  f         f  Tg         Ug  h         h  i         i  j         j  ,l         l  	n         ړ             D         F  z         |                                   R         T                                   *         ,  `         b             ̤         Τ             8         :  n         p             ڪ         ܪ             F         H  |         ~                                    T         V                      µ             ,                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               C!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 c
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     -;-    ;-M;-    M;-I-    I-5}-l    5}-S-    S--P    --    -b-h	    b-E-p
    E-S-
    S-۰-    ۰-E׳-L                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               E!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             ,         -  Y         Z                                                                                  6            S  D            q                                                                     K          L   x!         y!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        T#  $         $  %         %  &         &  2(         4(  j)         l)  *         *  +         +  -         -  J.         L.  /         /  0         0  1         1  *3         ,3  b4         d4  5         5  6         6  
8         8  B9         D9  z:         |:  ;         ;  <         <  ">         $>  Z?         \?  @         @  A         A  C         C  :D         <D  rE         tE  F         F  G         G  I         I  SJ         TJ  K         K  L         L  M         M  3O         4O  kP         lP  Q         Q  R         R  T         T  KU         LU  V         V  W         W  X         X  +Z         ,Z  c[         d[  \         \  ]         ]  _         _  C`         D`  {a         |a  b         b  c         c  #e         $e  [f          \f  [g          g  h          h  i          j  k         <k  sl         tl  m         m  n         n  p         p  Sq         Tq  r         r  s         s  t         t  3v         4v  kw         lw  x         x  y         y  {         {  K|         L|  }         }  ~         ~             +         ,  c         d             ӄ         Ԅ             C         D  {         |                        #         $  [         \             ˏ          ̏  ː                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               M!         ^!  1         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    t
         u
             ^                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     %
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-;-x    ;-I-|    I-|-\    |-A-    A--h    -b-    b--	    --h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               =!         @!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      :         ;  g	         h	  
         
           S                                              3          4            Q  5          n  R            o                                                           L   D!          y!  q"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            2         4  f         h                                   6         8  j         l  	         	  
         
             :         <  n         p                        
           >         @  r         t                                   B         D  v         x                                   F         H  z          |   !         !  "         "  $         $  J%         L%  ~&         &  '         '  (         (  *         *  N+         P+  ,         ,  -         -  .         .  0          0  R1         T1  2         2  3         3  4         4  "6         $6  V7         X7  8         8  9         9  :         :  &<         (<  Z=         \=  >         >  ?         ?  @         @  *B         ,B  ^C         `C  D         D  E         E  F         F  .H         0H  bI         dI  J         J  K         K  L          M  2N         4N  fO         hO  P         P  Q         Q  S         S  6T         8T  jU         lU  V         W  QX         SX  Y         Y  Z         Z  [         [  !]         #]  U^         W^  _         _  `         `  a         a  %c         'c  Yd         [d  e         e  f         f  g         g  )i         +i  ]j                      Ԅ  օ                      D  F          |  ~         T                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               <!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   *      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     -5-    5-e5-    e5-C-    C-Mw-l    Mw-U2-    U2--x    -b-	    b-E-
    E-S-
    S--|    -ֳ-    ֳ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               C!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         -            Z  L            y                                                          8	  
          U
  3         r                                   %         &  R                      =         >  j         k                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |        "  #         #  !%         #%  [&         ]&  '         '  (         (  	*         *  C+         E+  },         ,  -         -  .         .  +0         -0  e1         g1  2         2  3         3  5         5  M6         O6  7         7  8         8  9         9  5;         7;  o<         q<  =         =  >         >  @         @  WA         YA  B         B  C         C  E         E  ?F          AF  %G          aG  EH          H  eI          I  J          J  K         h  8i         :i  rj         tj  k         k  l         l   n         "n  Zo         \o  p         p  q         q  s         
s  Bt         Dt  |u         ~u  v         v  w         w  *y         ,y  dz         fz  {         {  |         |  ~         ~  L         N                                   4         6  n         p                                   V         X             ʌ         ̌             =         ?  w         y                        #         %  ]         _             ї         ӗ             E         G                                   -         /  g         i             ۢ         ݢ             O         Q             ç         ŧ             7         9  q         s                                 !  Y         [             Ͳ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               B!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    t
         u
             ^      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    t
         u
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     u
            0     -,-X     ,--     -/-    /-;-x    ;-I-|    I-|-\    |-A-    A--h    -b-    b--	    --h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ?!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      :         ;  g	         h	  
         
           S                                              3          4            Q  6          n  S            p                                                           L   C!          y!  p"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            3         4  g         h                                   7         8  k         l  	         	  
         
             ;         <  o         p                                   ?         @  s         t                                   C         D  w         x                                   G         H  {          |   !         !  "         "  $         $  K%         L%  &         &  '         '  (         (  *         *  O+         P+  ,         ,  -         -  .         .  0          0  S1         T1  2         2  3         3  4         4  #6         $6  W7         X7  8         8  9         9  :         :  '<         (<  [=         \=  >         >  ?         ?  @         @  +B         ,B  _C         `C  D         D  E         E  F         F  /H         0H  cI         dI  J         J  K         K  L          M  3N         4N  gO         hO  P         P  Q         Q  S         S  7T         8T  kU         lU  V         W  RX         SX  Y         Y  Z         Z  [         [  "]         #]  V^         W^  _         _  `         `  a         a  &c         'c  Zd         [d  e         e  f         f  g         g  *i         +i  ^j                                  $  &          \  ^                     T                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               8!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   )      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     -5-    5-e5-    e5-C-    C-Mw-l    Mw-U2-    U2--x    -b-	    b-E-
    E-S-
    S--|    -ֳ-    ֳ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               C!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         -            Z  K            x                                                          8	  
          U
  4         r                                   %         &  R                      =         >  j         k                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |        "  #         #  "%         #%  \&         ]&  '         '  (         (  
*         *  D+         E+  ~,         ,  -         -  .         .  ,0         -0  f1         g1  2         2  3         3  5         5  N6         O6  7         7  8         8  9         9  6;         7;  p<         q<  =         =  >         >  @         @  XA         YA  B         B  C         C  E         E  @F          K  L          M  M          !N  O          AO  %P          aP  EQ         h  9i         :i  sj         tj  k         k  l         l  !n         "n  [o         \o  p         p  q         q  	s         
s  Ct         Dt  }u         ~u  v         v  w         w  +y         ,y  ez         fz  {         {  |         |  ~         ~  M         N                                   5         6  o         p                                   W         X             ˌ         ̌             >         ?  x         y                        $         %  ^         _             җ         ӗ             F         G                                   .         /  h         i             ܢ         ݢ             P         Q             ħ         ŧ             8         9  r         s                                  !  Z         [             β                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               @!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    t
         u
             ^      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    t
         u
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     v
            0     -,-X     ,--     -M-    M-A;-x    A;-I-|    I-)}-\    )}--    --h    -ib-    ib-	-	    	--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               S!         V!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             ,         -  Y         Z                                  r  j                                              &            S  4          p  Q            n                                                         K          L   x!         y!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            0         2  b         d                                   *         ,  \         ^  	         	  
         
             $         &  V         X                                               P         R                                              J         L  |         ~                                   D          F   v!         x!  "         "  #          $  $          ,%  &          L&  3'          l'  S(          (  s)         F  G         G  &I         (I  XJ         ZJ  K         K  L         L  M         M   O         "O  RP         TP  Q         Q  R         R  S         S  U         U  LV         NV  ~W         W  X         X  Y         Y  [         [  F\         H\  x]         z]  ^         ^  _         _  a         a  @b         Bb  rc         tc  d         d  e         e  g         
g  :h         <h  li         ni  j         j  k         k  m         m  5n         6n  go         ho  p         p  q         q  r         r  /t         0t  au         bu  v         v  w         w  x         x  )z         *z  [{         \{  |         |  }         }  ~         ~  #         $  U         V                                              O         P                                              I         J  {                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               3!         :!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   *      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-A;-x    A;-I-|    I-)}-\    )}--    --h    -b-    b--n-	    -n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               B!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             ,         -  Y         Z                                                                                  6            S  E            r                                                                     K          L   x!         y!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ;         <  w         x                        +         ,  g         h             	         	             W         X                        
           F         H                                   6         8  r         t                        &         (  b         d                                     R!         T!  "         "  #         #  %          1  2          2  3          
4  4          :5  .6          j6  ^7         :H  uI         vI  J         J  K         K  )M         *M  eN         fN  O         O  P         P  R         R  US         VS  T         T  U         U  	W         
W  EX         FX  Y         Y  Z         Z  [         [  5]         6]  q^         r^  _         _  `         `  %b         &b  ac         bc  d         d  e         e  g         g  Qh         Rh  i         i  j         j  l         l  Am         Bm  |n         ~n  o         o  p         p  0r         2r  ls         ns  t         t  u         u   w         "w  \x         ^x  y         y  z         z  |         |  L}         N}  ~         ~                         <         >  x         z                        ,         .  h         j                                   X         Z             Џ         ҏ             H                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               C!         V!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     O
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -M-    M-A;-x    A;-I-|    I-)}-\    )}--    --h    -cb-    cb-	-	    	--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               Q!         V!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             ,         -  Y         Z                                  r  i                                              &            S  5          p  R            o                                                         K          L   x!         y!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            1         2  c         d                                   +         ,  ]         ^  	         	  
         
             %         &  W         X                                               Q         R                                              K         L  }         ~                                   E          F   w!         x!  "         "  #          )  *          *  +          +  ,          -  -          ,.  /         F  G         G  'I         (I  YJ         ZJ  K         K  L         L  M         M  !O         "O  SP         TP  Q         Q  R         R  S         S  U         U  MV         NV  W         W  X         X  Y         Y  [         [  G\         H\  y]         z]  ^         ^  _         _  a         a  Ab         Bb  sc         tc  d         d  e         e  	g         
g  ;h         <h  mi         ni  j         j  k         k  m         m  4n         6n  fo         ho  p         p  q         q  r         r  .t         0t  `u         bu  v         v  w         w  x         x  (z         *z  Z{         \{  |         |  }         }  ~         ~  "         $  T         V                                              N         P                                              H         J  z                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               1!         :!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   )      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -;-    ;-A;-x    A;-I-|    I-)}-\    )}--    --h    -b-    b--n-	    -n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               D!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             ,         -  Y         Z                                                                                  6            S  D            q                                                                     K          L   x!         y!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            :         <  v         x                        *         ,  f         h             	         	             V         X                                   G         H                                   7         8  s         t                        '         (  c         d                                     S!         T!  "         "  #         #  %          7  8          8  9          9  :          *;  <          Z<  N=         :H  tI         vI  J         J  K         K  (M         *M  dN         fN  O         O  P         P  R         R  TS         VS  T         T  U         U  W         
W  DX         FX  Y         Y  Z         Z  [         [  4]         6]  p^         r^  _         _  `         `  $b         &b  `c         bc  d         d  e         e  g         g  Ph         Rh  i         i  j         j  l         l  @m         Bm  }n         ~n  o         o  p         p  1r         2r  ms         ns  t         t  u         u  !w         "w  ]x         ^x  y         y  z         z  |         |  M}         N}  ~         ~                        =         >  y         z                        -         .  i         j                                   Y         Z             я         ҏ             I                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               C!         V!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     N
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -G-    G-;-    ;-I-    I-|-d    |-
-    
-W-0    W--    -b-H	    b-c-P
    c-U-
    U-9׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               N!         X!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      :         ;  g	         h	  
         
           S                                              3          4            Q  5          n  R            o                                                           L   D!          y!  q"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q   w  operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       \            .         0  ^         `                                               N         P  ~	         	  
         
                        >         @  n         p                                    .         0  ^         `                                               N         P  ~                                              >!         @!  n"         %  &         &  ((         *(  X)         Z)  *         *  +         +  ,         ,  .         .  H/         J/  x0         z0  1         1  2         2  4         
4  85         :5  h6         j6  7         7  8         8  9         9  (;         *;  X<         Z<  =         =  >         >  ?         ?  A         A  HB         JB  xC         zC  D         D  E         E  G         
G  8H         h   j         j  0k         2k  `l         bl  m         m  n         n  o         o   q         "q  Pr         Rr  s         s  t          t  u          u  v          w  	x          Bx  9y          ry  iz         z  {         {   }         }  0~         2~  `         b                                              L         N  {         }             ۈ         ݈             ;                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               B!         X!  1         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 s
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-_5-x    _5-C-|    C-Gw-\    Gw-51-    51--h    -b-    b--n-	    -n--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               B!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         -            Z  L            y                                                          8	  
          U
  3         r                                   %         &  R                      =         >  j         k                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            =         >  {         |                        5         6  s         t             	         	  -         .  k         l                        %         &  c         d                                   [         \                                   S         T                                   K          L   !         !  "         "  $         $  C%         G  H         H  0J         2J  nK         pK  L         L  M         M  (O         *O  fP         hP  Q         Q  R         R   T         "T  ^U         `U  V         V  W         W  Y         Y  VZ         XZ  [         [  \         \  ^         ^  N_         P_  `         `  a         a  c         
c  Fd         Hd  e         e  f         f   h         h  >i         @i  |j         ~j  k         k  l         n  o         o  *q         ,q  hr         jr  s         s  t         t  "v         $v  `w         bw  x         x  y         y  {         {  X|         Z|  }         }  ~         ~             P         R             ̃         ΃  
           H         J             Ĉ         ƈ             @         B  ~                                 8         :  v         x                                               4  8          t  x                               >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               B!         T!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     O
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -A-    A-;-    ;-I-    I-|-d    |-
-    
-W-0    W--    -b-H	    b-c-P
    c-U-
    U-9׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               K!         V!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      :         ;  g	         h	  
         
           S                                              3          4            Q  6          n  S            p                                                           L   C!          y!  p"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q   w  operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       \            /         0  _         `                                               O         P  	         	  
         
                        ?         @  o         p                                    /         0  _         `                                               O         P                                                ?!         @!  o"         %  &         &  )(         *(  Y)         Z)  *         *  +         +  ,         ,  .         .  I/         J/  y0         z0  1         1  2         2  	4         
4  95         :5  i6         j6  7         7  8         8  9         9  );         *;  Y<         Z<  =         =  >         >  ?         ?  A         A  IB         JB  yC         zC  D         D  E         E  	G         
G  9H         h  j         j  1k         2k  al         bl  m         m  n         n  o         o  !q         "q  Qr         Rr  s         s  t         t  u         u  w         w  Ax         Bx  qy         ry  z          z  {          {  |          }  }          2~  )          b  Y                                            M         N  |         }             ܈         ݈             <                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               >!         T!  1         1  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 s
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-_5-x    _5-C-|    C-Gw-\    Gw-w7-    w7--h    -b-    b--n-	    -n--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               D!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         -            Z  K            x                                                          8	  
          U
  4         r                                   %         &  R                      =         >  j         k                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            <         >  z         |                        4         6  r         t             	         	  ,         .  j         l                        $         &  b         d                                   Z         \                                   R         T                                   J          L   !         !  "         "  $         $  B%         G  H         H  1J         2J  oK         pK  L         L  M         M  )O         *O  gP         hP  Q         Q  R         R  !T         "T  _U         `U  V         V  W         W  Y         Y  WZ         XZ  [         [  \         \  ^         ^  O_         P_  `         `  a         a  	c         
c  Gd         Hd  e         e  f         f  h         h  ?i         @i  }j         ~j  k         k  l         n  o         o  +q         ,q  ir         jr  s         s  t         t  #v         $v  aw         bw  x         x  y         y  {         {  Y|         Z|  }         }  ~         ~             Q         R             ̓         ΃             I         J             ň         ƈ             A         B                                   9         :  w         x                                   4  8          t  x                                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               @!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     N
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -M-    M-;;-    ;;-I-    I-#}-d    #}-!-    !--p    -b-    b-9-
    9-G-
    G--t    -3׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               P!         `!   2         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             ,         -  Y         Z                                  r  j                                              &            S  4          p  Q            n                                                         K          L   x!         y!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                      @  G                                              %  &         &  '         '  )         
)  6*         8*  d+         f+  ,         ,  -         -  .         .  0         0  J1         L1  x2         z2  3         3  4         4  6         6  07         27  ^8         `8  9         9  :         :  ;         ;  =         =  D>         F>  r?         t?  @         @  A         A  B         B  *D         ,D  XE         ZE  F         F  G         G  H         H  J         J  >K         @K  lL         nL  M         M  N         N  O         O  %Q         &Q  SR         TR  S         S  T         T  U         U  W         W  9X         :X  gY         hY  Z         Z  [         [  \         \  ^          ^  M_         N_  {`         |`  a         a  b         b  d         d  3e         4e  af         bf  g         g  h         h  i         i  k         k  Gl         Hl  um         vm  n         n  o         o  p          q  ,r         .r  Zs         \s  t         t  u         u  v         v  x         x  @y         By  nz         pz  {         {  |         |  }         }  &         (  T         V                        ރ                      :         <  h         j             ĉ         Ɖ                       "  N         P  |         ~                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               S!         j!  2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    t
         u
             ^      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     R
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -;-    ;-;-x    ;-YJ-|    YJ-}-\    }-,-    ,--h    -Qb-    Qb-ӂ-	    ӂ--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               D!         X!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                               	  5	         6	  b
         c
                                   G         H  t         u                                                Y  [                                                            *!  "          G"  (#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            -         .  [         \                                              A         B  o	         p	  
         
                        '         (  U         V                                              ;         <  i         j                                   !         "  O         P  }         ~                        !         !  5"         6"  c#          d#  D$          $  `%         %  &         &  '         '  %)         &)  S*         T*  +         +  ,         ,  -         -  /         /  90         :0  g1         h1  2         2  3         3  4         4  6          6  M7         N7  {8         |8  9         9  :         :  <         <  3=         4=  a>         b>  ?         ?  @         @  A         A  C         C  GD         HD  uE         vE  F         F  G         G  H         TI  J         J  K          m  m          n  n          ;o  p           ?         A  m         o             ɐ         ː             %         '  S         U                        ݗ         ߗ             9         ;  g         i             Ý         ŝ                      !  M         O  {         }             פ         ٤             3         5  a         c                                                      >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               '!         6!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -G-    G-;;-    ;;-I-    I-#}-d    #}-'-    '--p    -b-    b-9-
    9-G-
    G--t    -3׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               L!         \!  1         
2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             ,         -  Y         Z                                  r  i                                              &            S  5          p  R            o                                                         K          L   x!         y!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         @  G              !          !  "           #  $          @$  G%         %  &         &  '         '  	)         
)  7*         8*  e+         f+  ,         ,  -         -  .         .  0         0  K1         L1  y2         z2  3         3  4         4  6         6  17         27  _8         `8  9         9  :         :  ;         ;  =         =  E>         F>  s?         t?  @         @  A         A  B         B  +D         ,D  YE         ZE  F         F  G         G  H         H  J         J  ?K         @K  mL         nL  M         M  N         N  O         O  $Q         &Q  RR         TR  S         S  T         T  U         U  
W         W  8X         :X  fY         hY  Z         Z  [         [  \         \  ^          ^  L_         N_  z`         |`  a         a  b         b  d         d  2e         4e  `f         bf  g         g  h         h  i         i  k         k  Fl         Hl  tm         vm  n         n  o         o  p          q  -r         .r  [s         \s  t         t  u         u  v         v  x         x  Ay         By  oz         pz  {         {  |         |  }         }  '         (  U         V                        ߃                      ;         <  i         j             ŉ         Ɖ             !         "  O         P  }         ~                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               S!         j!  2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    t
         u
             ^      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     Q
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-;-x    ;-_J-|    _J-}-\    }-,-    ,--h    -Wb-    Wb-ӂ-	    ӂ--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               B!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 :	         ;	  g
         h
                                    L         M  y         z                                    !  "          ^  _                                                            /!  "          L"  .#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ,         .  Z         \                                              @         B  n	         p	  
         
                        &         (  T         V                                              :         <  h         j                                             "  N         P  |         ~                        !         !  4"         6"  b#         %  &         &  '         '  $)         &)  R*         T*  +         +  ,         ,  -         -  
/         /  80         :0  f1         h1  2         2  3         3  4         4  6          6  L7         N7  z8         |8  9         9  :         :  <         <  2=         4=  `>         b>  ?         ?  @         @  A         A  C         C  FD         HD  tE         vE  F         F  G         G  H         TI  J         J  K          Wp  7q          sq  Sr          r  os          s  t          t  u           @         A  n         o             ʐ         ː             &         '  T         U                        ޗ         ߗ             :         ;  h         i             ĝ         ŝ                       !  N         O  |         }             ؤ         ٤             4         5  b         c                                                      >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               )!         :!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -;-    ;-C6-x    C6-D-|    D-+x-\    +x--    --h    -b-    b-m-	    m--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               H!         X!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     :            W  8          t  U                                    	          H	  D
          
                                   H         I  u         v                         -         .  Z         [                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            <         >  z         |                        4         6  r         t             	         	  ,         .  j         l                        $         &  b         d                                   Z         \                                   R         T                                   J          L   !         !  "         "  $         $  B%          D%  8&          t&  h'          '  (          (  )          *  *         H  !J         "J  _K         `K  L         L  M         M  O         O  WP         XP  Q         Q  R         R  T         T  OU         PU  V         V  W         W  	Y         
Y  GZ         HZ  [         [  \         \  ^         ^  ?_         @_  }`         ~`  a         a  b         b  7d         8d  ue         ve  f         f  g         g  /i         0i  mj         nj  k         k  l         l  'n         o  q         q  Er         Fr  s         s  t         t  u          v  =w         >w  {x         |x  y         y  z         z  5|         6|  s}         t}  ~         ~             -         .  k         l                        %         &  c         d             ߉                      [         \             ׎         ؎             S         T             ϓ         Г                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               H!         X!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     +
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-y-     y--    -5;-x    5;-I-|    I-}-\    }-{-    {--h    -9b-    9b-ӂ-	    ӂ--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               6!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             ,         -  Y         Z                                                                                  6            S  E            r                                                                     K          L   x!         y!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            .         0  ^         `                                               N         P  ~	         	  
         
                        >         @  n         p                                    .         0  ^         `                                               N         P  ~                                              >!         @!  n"         p"  #         D%  s&         t&  '         '  (         (  *         *  3+         4+  c,         d,  -         -  .         .  /         /  #1         $1  S2         T2  3         3  4         4  5         5  7         7  C8         D8  s9         t9  :         :  ;         ;  =         =  3>         4>  c?         d?  @         @  A         A  B         B  #D         $D  SE         TE  F         F  G         G  H          X  X          .Y  Z          ZZ  J[          [  v\          \  ]         ir  s         s  t         t  u         u  'w         )w  Wx         Yx  y         y  z         z  {         {  }         }  G~         I~  w         y             ׁ         ف           	  7         9  g         i             Ǉ         ɇ             '         )  W         Y                                              G         I  w         y             ה         ٔ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                                !         *!  1         1  z=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-C6-x    C6-D-|    D-+x-\    +x--    --h    -b-    b-m-	    m--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               D!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     :            W  9          t  V                                    	          H	  C
          
                                   H         I  u         v                         -         .  Z         [                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            =         >  {         |                        5         6  s         t             	         	  -         .  k         l                        %         &  c         d                                   [         \                                   S         T                                   K          L   !         !  "         "  $         $  C%          4+  (,          d,  X-          -  .          .  /          /  0         H   J         "J  ^K         `K  L         L  M         M  O         O  VP         XP  Q         Q  R         R  T         T  NU         PU  V         V  W         W  Y         
Y  FZ         HZ  [         [  \         \   ^         ^  >_         @_  |`         ~`  a         a  b         b  6d         8d  te         ve  f         f  g         g  .i         0i  lj         nj  k         k  l         l  &n         o  q         q  Dr         Fr  s         s  t         t  u          v  <w         >w  zx         |x  y         y  z         z  4|         6|  r}         t}  ~         ~             ,         .  j         l                        $         &  b         d             މ                      Z         \             ֎         ؎             R         T             Γ         Г                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               F!         T!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     *
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-s-     s--    -5;-x    5;-I-|    I-}-\    }-{-    {--h    -?b-    ?b-ӂ-	    ӂ--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               4!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             ,         -  Y         Z                                                                                  6            S  D            q                                                                     K          L   x!         y!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            /         0  _         `                                               O         P  	         	  
         
                        ?         @  o         p                                    /         0  _         `                                               O         P                                                ?!         @!  o"         p"  #         D%  r&         t&  '         '  (         (  *         *  2+         4+  b,         d,  -         -  .         .  /         /  "1         $1  R2         T2  3         3  4         4  5         5  7         7  B8         D8  r9         t9  :         :  ;         ;  =         =  2>         4>  b?         d?  @         @  A         A  B         B  "D         $D  RE         TE  F         F  G         G  H          ]  ^          
_  _          6`  &a          ba  Rb          b  ~c         ir  s         s  t         t  u         u  (w         )w  Xx         Yx  y         y  z         z  {         {  }         }  H~         I~  x         y             ؁         ف           	  8         9  h         i             ȇ         ɇ             (         )  X         Y                                              H         I  x         y             ؔ         ٔ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               "!         .!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-:-x    :-I-|    I-|-\    |-? -    ? --h    -b-    b-_x-	    _x--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               D!         X!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      :         ;  g	         h	  
         
           S                                              3          4            Q  5          n  R            o                                                           L   D!          y!  q"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            :         <  v         x                        *         ,  f         h             	         	             V         X                        
           F         H                                   6         8  r         t                        &         (  b         d                                     R!         T!  "         "  #         #  %         %  B&         D&  ~'         '  (         (  )         )  2+         4+  n,         p,  -         -  .         .  "0         $0  ^1         `1  2         2  3         3  5         5  N6         P6  7         7  8         8  :         :  >;         @;  z<         |<  =         =  >         >  .@         0@  jA         lA  B         B  C         C  E          E  ZF         \F  G         G  H         H  J         J  WK         XK  L         L  M         M  O         O  GP         HP  Q         Q  R         R  S         S  7U         8U  sV         tV  W         W  X         X  'Z         (Z  c[         d[  \         \  ]         ]  _         _  S`         T`  a         a  b         b  d         d  Ce         De  f         f  g         g  h         h  3j         4j  ok         pk  l         l  m         m  #o          *{  |          V|  I}          }  u~          ~              ̀                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ]!         ^!  2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 c
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     --    -;-x    ;-MJ-|    MJ-}-\    }--    --P    -{b-    {b-c-	    c-+-D
    +--׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               C!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                               	  5	         6	  b
         c
                                   G         H  t         u                                                Y  [                                                            *!  "          G"  (#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            1         2  c         d                                   +         ,  ]         ^  	         	  
         
             %         &  W         X                                               Q         R                                              K         L  }         ~                                   E          F   w!         x!  "         "  #         #  %         %  ?&         @&  q'         r'  (         (  )         )  +         +  9,         :,  k-         l-  .         .  /         /  1         1  32         42  e3         f3  4         4  5         5  6         6  -8         .8  _9         `9  :         :  ;         ;  <         <  '>         (>  Y?         Z?  @         @  A         A  B         B  !D         "D  SE         TE  F         F  G         G  H         H  J          b  c          d  e          De  Df          f  g          g  h         ir  s         s  t         t  u         u  /w         1w  ax         cx  y         y  z         z  {         {  )}         +}  [~         ]~                                   #         %  U         W                                              O         Q                                              I                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               G!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 g
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-:-x    :-I-|    I-|-\    |--    --h    -b-    b-_x-	    _x--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               B!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      :         ;  g	         h	  
         
           S                                              3          4            Q  6          n  S            p                                                           L   C!          y!  p"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ;         <  w         x                        +         ,  g         h             	         	             W         X                                   G         H                                   7         8  s         t                        '         (  c         d                                     S!         T!  "         "  #         #  %         %  C&         D&  '         '  (         (  )         )  3+         4+  o,         p,  -         -  .         .  #0         $0  _1         `1  2         2  3         3  5         5  O6         P6  7         7  8         8  :         :  ?;         @;  {<         |<  =         =  >         >  /@         0@  kA         lA  B         B  C         C  E          E  [F         \F  G         G  H         H  J         J  VK         XK  L         L  M         M  
O         O  FP         HP  Q         Q  R         R  S         S  6U         8U  rV         tV  W         W  X         X  &Z         (Z  b[         d[  \         \  ]         ]  _         _  R`         T`  a         a  b         b  d         d  Be         De  ~f         f  g         g  h         h  2j         4j  nk         pk  l         l  m         m  "o                      2  %          ^  Q            }                               >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ]!         ^!  2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 c
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-y-     y--    -;-x    ;-SJ-|    SJ-}-\    }--    --P    -{b-    {b-c-	    c-+-D
    +--׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               @!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 :	         ;	  g
         h
                                    L         M  y         z                                    !  "          ^  _                                                            /!  "          L"  .#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            0         2  b         d                                   *         ,  \         ^  	         	  
         
             $         &  V         X                                               P         R                                              J         L  |         ~                                   D          F   v!         x!  "         "  #         #  %         %  >&         @&  p'         r'  (         (  )         )  +         +  8,         :,  j-         l-  .         .  /         /   1         1  22         42  d3         f3  4         4  5         5  6         6  ,8         .8  ^9         `9  :         :  ;         ;  <         <  &>         (>  X?         Z?  @         @  A         A  B         B   D         "D  RE         TE  F         F  G         G  H         H  J          h  i          4j  4k          pk  pl          l  m          m  n         ir  s         s  t         t  u         u  0w         1w  bx         cx  y         y  z         z  {         {  *}         +}  \~         ]~                                   $         %  V         W                                              P         Q                                              J                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               G!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 g
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -u,-X     u,-+-     +--    -[6-x    [6-D-|    D-Cx-\    Cx-1-    1--h    -ub-    ub-m-	    m--h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     R                       D!  1          1  h=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     6            Q  .          l  I                                    		          D	  H
          
                                    N         O  }         ~                        E         F  t         u                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            4         6  j         l                                   B         D  x         z  	         	  
         
             P         R                                   (         *  ^         `                                    6         8  l         n                                   D         F  z         |               !         !  #         #  R$          ;  p<          <  =          =  >          >  ?           @  A          I  UJ         VJ  K         K  L         L  M         M  -O         .O  cP         dP  Q         Q  R         R  T         T  ;U         <U  qV         rV  W         W  X         X  Z         Z  I[         J[  \         \  ]         ]  ^         ^  !`         "`  Wa         Xa  b         b  c         c  d         d  /f         0f  eg         fg  h         h  i         i  k         k  =l         >l  sm         tm  n         n  o         o  q         q  Kr         Lr  s         s  t         t  u         u  #w         $w  Yx         Zx  y         y  z         z  {         {  1}         2}  g~         h~             Ӏ         Ԁ  	         
  ?         @  u         v                                   M         N                                   %         &  [         \             Ǒ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               G!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     +
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,-U-     U--    -;-x    ;-I-|    I-|-\    |--    --h    -Qb-    Qb--	    --h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     a            !          V!  1          2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             .         /  ]         ^                                                                                  .            I  ?          t  j                                                                     3          4   b!         c!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            6         8  n         p                                   N         P             	         	  
         
  .         0  f         h                                   F         H  ~                                 &         (  ^         `                                   >         @  v         x                          "          "  V#         X#  $         $  %         %  &          '  7(         8(  o)         p)  *         *  +         +  -         -  O.         P.  /         /  0         0  1         1  /3         03  g4         h4  5         5  6         6  8         8  G9         H9  :         :  ;         ;  <         <  '>         (>  _?         `?  @         @  A         A  C         C  ?D         @D  wE         xE  F         F  G         G  I          <U  4V          rV  jW          W  X          X  Y          Z  [         o   q         q  8r         :r  ps         rs  t         t  u         u  w         w  Px         Rx  y         y  z         z  {         {  0}         2}  h~         j~             ؀         ڀ             H         J                                   (         *  `         b             Ћ         ҋ           
  @         B  x         z                                  "  X                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               5!         6!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -u,-X     u,-+-     +--    -[6-x    [6-D-|    D-Cx-\    Cx-1-    1--h    -ib-    ib-m-	    m--h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     S                       D!  1          1  i=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     6            Q  /          l  J                                    	          D	  G
          
                                    N         O  }         ~                        E         F  t         u                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            5         6  k         l                                   C         D  y         z  	         	  
         
             Q         R                                   )         *  _         `                                   7         8  m         n                                   E         F  {         |               !         !  #         #  S$          FA  .B          lB  TC          C  zD          D  E          E  F          I  TJ         VJ  K         K  L         L  M         M  ,O         .O  bP         dP  Q         Q  R         R  T         T  :U         <U  pV         rV  W         W  X         X  Z         Z  H[         J[  ~\         \  ]         ]  ^         ^   `         "`  Va         Xa  b         b  c         c  d         d  .f         0f  dg         fg  h         h  i         i  k         k  <l         >l  rm         tm  n         n  o         o  q         q  Jr         Lr  s         s  t         t  u         u  "w         $w  Xx         Zx  y         y  z         z  {         {  0}         2}  f~         h~             Ҁ         Ԁ           
  >         @  t         v                                   L         N                                   $         &  Z         \             Ƒ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               C!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     *
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,-I-     I--    -;-x    ;-I-|    I-|-\    |--    --h    -Wb-    Wb--	    --h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     `            	!          R!  1          1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             .         /  ]         ^                                                                                  .            I  >          t  i                                                                     3          4   b!         c!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            7         8  o         p                                   O         P             	         	  
         
  /         0  g         h                                   G         H                                   '         (  _         `                                   ?         @  w         x                          "          "  W#         X#  $         $  %         %  &          '  6(         8(  n)         p)  *         *  +         +  -         -  N.         P.  /         /  0         0  1         1  .3         03  f4         h4  5         5  6         6  8         8  F9         H9  ~:         :  ;         ;  <         <  &>         (>  ^?         `?  @         @  A         A  C         C  >D         @D  vE         xE  F         F  G         G  I          J[  B\          \  x]          ]  ^          ^  _          "`  a         o  q         q  9r         :r  qs         rs  t         t  u         u  w         w  Qx         Rx  y         y  z         z  {         {  1}         2}  i~         j~             ـ         ڀ             I         J                                   )         *  a         b             ы         ҋ  	         
  A         B  y         z                        !         "  Y                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               9!         :!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,-=-     =--    -;-x    ;-I-|    I-|-\    |-G-    G--h    -ib-    ib-Sx-	    Sx--h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     V                       L!  1          1  x=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      4         5  c	         d	  
         
           ]                                              G          H  $          g  C            b                                                          9  %           d   P!          !  {"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            2         4  f         h                                   6         8  j         l  	         	  
         
             :         <  n         p                        
           >         @  r         t                                   B         D  v         x                                   F         H  z          |   !         !  "         "  $         $  J%         L%  ~&         &  '         '  (         (  *         *  N+         P+  ,         ,  -         -  .         .  0          0  R1         T1  2         2  3         3  4         4  "6         $6  V7         X7  8         8  9         9  :         :  &<         (<  Z=         \=  >         >  ?         ?  @         @  *B         ,B  ^C         `C  D         D  E         E  F         F  .H         ^H  I         I  J         J  K         K  -M         .M  aN         bN  O         O  P         P  Q         Q  1S         2S  eT         fT  U         U  V         V  X         X  5Y         6Y  iZ         jZ  [         [  \         \  ^         ^  9_         :_  m`         n`  a         a  b         b  	d         
d  =e         >e  qf         rf  g         g  h         h  j         j  Ak         Bk  ul          o  p          q  r          Pr  Ps          s  t          t  u                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               @!         B!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 a
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,-O-     O--    -;-x    ;-MJ-|    MJ-}-\    }-3-    3--P    -b-    b-c-	    c--D
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     a            	!          P!  1          1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 	         	  L
         M
  {                                 C         D  r         s                                                 _  Y                                                            .!  "          I"  &#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            9         :  s         t                        !         "  [         \             	         	  	         
  C         D  }         ~                        +         ,  e         f                                   M         N                                   5         6  o         p                        !         !  W"         X"  #         #  $         $  &         &  ?'         @'  y(         z(  )         )  *         *  ',         (,  a-         b-  .         .  /         /  1         1  I2         J2  3         3  4         4  5         5  17         27  k8         l8  9         9  :         :  <         <  S=         T=  >         >  ?         ?  A         A  ;B         <B  uC         vC  D         D  E         E  #G         $G  \H         vl  m         m  n         n  "p         $p  \q         ^q  r         r  s         s  
u         u  Dv         Fv  ~w         w  x         x  y         y  ,{         .{  f|         h|  }         }  ~         ~             N         P                      ă             6         8  p         r                                    X         Z             ̎                      ٮ  ޯ            !          _  d                               >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               J!         T!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 e
                     
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,-7-     7--    -;-x    ;-I-|    I-|-\    |--    --h    -ob-    ob-Mx-	    Mx--h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     U                       H!  1          1  q=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      4         5  c	         d	  
         
           ]                                              G          H  %          g  D            c                                                          9  $           d   O!          !  z"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            3         4  g         h                                   7         8  k         l  	         	  
         
             ;         <  o         p                                   ?         @  s         t                                   C         D  w         x                                   G         H  {          |   !         !  "         "  $         $  K%         L%  &         &  '         '  (         (  *         *  O+         P+  ,         ,  -         -  .         .  0          0  S1         T1  2         2  3         3  4         4  #6         $6  W7         X7  8         8  9         9  :         :  '<         (<  [=         \=  >         >  ?         ?  @         @  +B         ,B  _C         `C  D         D  E         E  F         F  /H         ^H  I         I  J         J  K         K  ,M         .M  `N         bN  O         O  P         P  Q         Q  0S         2S  dT         fT  U         U  V         V   X         X  4Y         6Y  hZ         jZ  [         [  \         \  ^         ^  8_         :_  l`         n`  a         a  b         b  d         
d  <e         >e  pf         rf  g         g  h         h  j         j  @k         Bk  tl          v  w          \w  \x          x  y          y  z          %{  %|                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               B!         F!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 a
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,-I-     I--    -;-x    ;-SJ-|    SJ-}-\    }-c9-    c9--P    -b-    b-c-	    c--D
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     `            !          N!  1          1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 "	         #	  Q
         R
                                   H         I  w         x                                    %            d  ]                                                            3!  "          N"  ,#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            8         :  r         t                                  "  Z         \             	         	           
  B         D  |         ~                        *         ,  d         f                                   L         N                                   4         6  n         p                        !         !  V"         X"  #         #  $         $  &         &  >'         @'  x(         z(  )         )  *         *  &,         (,  `-         b-  .         .  /         /  1         1  H2         J2  3         3  4         4  5         5  07         27  j8         l8  9         9  :         :  <         <  R=         T=  >         >  ?         ?   A         A  :B         <B  tC         vC  D         D  E         E  "G         $G  ]H         vl  m         m  n         n  #p         $p  ]q         ^q  r         r  s         s  u         u  Ev         Fv  w         w  x         x  y         y  -{         .{  g|         h|  }         }  ~         ~             O         P             Ã         ă             7         8  q         r                                    Y         Z             ͎                      (  -          k  p                                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               L!         T!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 e
                     
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,-m-     m--    -a6-x    a6-D-|    D-Ix-\    Ix--    -k-h    k-	b-    	b-Sx-	    Sx--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               /!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     6            Q  .          l  I                                    		          D	  H
          
                                    N         O  }         ~                        E         F  t         u                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                 $  (          g  k                     $  %         %  '         '  8(         :(  j)         l)  *         *  +         +   -         -  2.         4.  d/         f/  0         0  1         1  2         2  ,4         .4  ^5         `5  6         6  7         7  8         8  &:         (:  X;         Z;  <         <  =         =  >         >   @         "@  RA         TA  B         B  C         C  D         D  F         F  LG         NG  ~H         H  I         I  K         K  BL         DL  tM         vM  N         N  O         O  
Q         Q  <R         >R  nS         pS  T         T  U         U  W         W  7X         8X  iY         jY  Z         Z  [         [  \          ]  1^         2^  c_         d_  `         `  a         a  b         b  +d         ,d  ]e         ^e  f         f  g         g  h         h  %j         &j  Wk         Xk  l         l  m         m  n         n  p          p  Qq         Rq  r         r  s         s  t         t  v         v  Kw         Lw  }x         ~x  y         y  z         z  |         |  E}         F}  w~         x~             ۀ         ܀             ?         @  q         r             Ն         ֆ             9         :  k         l             ό         Ќ             3         d                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     P                       4!  1          1  N=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 a
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     --    -;-    ;-I-    I-}-d    }--    --p    -]b-    ]b-3-
    3-A-
    A--t    --׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               =!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             .         /  ]         ^                                                                                  .            I  ?          t  j                                                                     3          4   b!         c!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                        *            T  C          ~  m                     T#  $         $  %         %  '         '  C(         D(  )         )  *         *  +         +  3-         4-  o.         p.  /         /  0         0  #2         $2  _3         `3  4         4  5         5  7         7  O8         P8  9         9  :         :  <         <  ?=         @=  {>         |>  ?         ?  @         @  /B         0B  kC         lC  D         D  E         E  G          G  [H         \H  I         J  K         K  8M         :M  tN         vN  O         O  P         P  (R         *R  dS         fS  T         T  U         U  W         W  TX         VX  Y         Y  Z         Z  \         
\  D]         F]  ^         ^  _         _  `         `  4b         6b  pc         rc  d         d  e         e  $g         &g  `h         bh  i         i  j         j  l         l  Pm         Rm  n         n  o         o  q         q  @r         Br  |s         ~s  t         t  u         u  0w         2w  lx         nx  y         y  z         z   |         "|  \}         ^}  ~         ~                        L         N             Ą         Ƅ              <         >  x         z                        ,         .  h         j                                   X         Z                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     a            !          b!  1          2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    v
         w
             d      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                      
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-g-     g--    -a6-x    a6-D-|    D-Ix-\    Ix--    -q-h    q-b-    b-Sx-	    Sx--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               -!         @!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     6            Q  /          l  J                                    	          D	  G
          
                                    N         O  }         ~                        E         F  t         u                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                     0  4          s  w                                 $  %         %  '         '  9(         :(  k)         l)  *         *  +         +  -         -  3.         4.  e/         f/  0         0  1         1  2         2  -4         .4  _5         `5  6         6  7         7  8         8  ':         (:  Y;         Z;  <         <  =         =  >         >  !@         "@  SA         TA  B         B  C         C  D         D  F         F  MG         NG  H         H  I         I  K         K  CL         DL  uM         vM  N         N  O         O  Q         Q  =R         >R  oS         pS  T         T  U         U  W         W  6X         8X  hY         jY  Z         Z  [         [  \          ]  0^         2^  b_         d_  `         `  a         a  b         b  *d         ,d  \e         ^e  f         f  g         g  h         h  $j         &j  Vk         Xk  l         l  m         m  n         n  p          p  Pq         Rq  r         r  s         s  t         t  v         v  Jw         Lw  |x         ~x  y         y  z         z  |         |  D}         F}  v~         x~             ڀ         ܀             >         @  p         r             Ԇ         ֆ             8         :  j         l             Ό         Ќ              2         d                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     Q                       8!  1          1  U=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 a
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     --    -;-    ;-I-    I-}-d    }--    --p    -]b-    ]b-3-
    3-A-
    A--t    --׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ;!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             .         /  ]         ^                                                                                  .            I  >          t  i                                                                     3          4   b!         c!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                 &  	          P	  ?
          z
  i         T#  $         $  %         %  '         '  B(         D(  ~)         )  *         *  +         +  2-         4-  n.         p.  /         /  0         0  "2         $2  ^3         `3  4         4  5         5  7         7  N8         P8  9         9  :         :  <         <  >=         @=  z>         |>  ?         ?  @         @  .B         0B  jC         lC  D         D  E         E  G          G  ZH         \H  I         J  K         K  9M         :M  uN         vN  O         O  P         P  )R         *R  eS         fS  T         T  U         U  W         W  UX         VX  Y         Y  Z         Z  	\         
\  E]         F]  ^         ^  _         _  `         `  5b         6b  qc         rc  d         d  e         e  %g         &g  ah         bh  i         i  j         j  l         l  Qm         Rm  n         n  o         o  q         q  Ar         Br  }s         ~s  t         t  u         u  1w         2w  mx         nx  y         y  z         z  !|         "|  ]}         ^}  ~         ~                        M         N             ń         Ƅ             =         >  y         z                        -         .  i         j                                   Y         Z                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     `            !          b!  1          2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    v
         w
             d      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                      
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-[-     [--    -;-x    ;-I-|    I-|-\    |-1-    1-}-h    }-b-    b--	    --h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               &!         8!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      4         5  c	         d	  
         
           ]                                              G          H  $          g  C            b                                                          9  %           d   P!          !  {"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            .         0  ^         `                                               N         P  ~	         	  
         
                        >         @  n         p                                    .         0  ^         `                                               N         P  ~                                              >!         @!  n"         p"  #         #  $         $  %          &  .'         0'  ^(         `(  )         )  *         *  +         +  -          -  N.         P.  ~/         /  0         0  1         1  3         3  >4         @4  n5         p5  6         6  7         7  8          9  .:         0:  ^;         `;  <         <  =         =  >         >  @          @  NA         PA  ~B         B  C         C  D         D  F         F  >G         @G  nH         pH  I         I  J          b  c          Bc  ,d          ld  Ve          e  f          f  g         d                      Ĕ             "         $  R         T                                              B         D  r         t             ҡ         ԡ             2         4  b         d             §         ħ             "         $  R         T                                              B         D  r                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     M                       @!  1          1  ]=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     -)-    )-w5-    w5-D-    D-_w-l    _w-A4-    A4--x    -b-	    b-9-
    9-G-
    G--|    -ճ-    ճ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ?!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         +  !          V  L            w                                                          4	  
          S
  9         r                                   -         .  \           &         '  U         V                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |        (#  c$          /  0          0  1          1  2          3  3          34  5         CF  ~G         G  H         H  I         I  8K         :K  vL         xL  M         M  N         N  0P         2P  nQ         pQ  R         R  S         S  (U         *U  fV         hV  W         W  X         X   Z         "Z  ^[         `[  \         \  ]         ]  _         _  V`         X`  a         a  b         b  d         d  Ne         Pe  f         f  g         g  i         
i  Fj         Hj  k         k  l         l   n         n  >o         @o  |p         ~p  q         q  r         r  6t         8t  tu         vu  v         v  w         w  .y         0y  lz         nz  {         {  |         |  &~         (~  d         f                                    \         ^             ؆         چ             T         V             Ћ         ҋ             L         N             Ȑ         ʐ             C         E                                   9         ;  w         y                        1         3  o         q                        )         +  g         i                        !         #  _         a             ۪         ݪ             W         Y             ӯ         կ             O         Q             ˴                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               @!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    v
         w
             d      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    v
         w
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            0     -,-X     ,-U-     U--    -;-x    ;-I-|    I-|-\    |-1-    1-w-h    w-b-    b--	    --h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               %!         6!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      4         5  c	         d	  
         
           ]                                              G          H  %          g  D            c                                                          9  $           d   O!          !  z"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            /         0  _         `                                               O         P  	         	  
         
                        ?         @  o         p                                    /         0  _         `                                               O         P                                                ?!         @!  o"         p"  #         #  $         $  %          &  /'         0'  _(         `(  )         )  *         *  +         +  -          -  O.         P.  /         /  0         0  1         1  3         3  ?4         @4  o5         p5  6         6  7         7  8          9  /:         0:  _;         `;  <         <  =         =  >         >  @          @  OA         PA  B         B  C         C  D         D  F         F  ?G         @G  oH         pH  I         I  J          g  h          i  i          >j  (k          hk  Rl          l  |m         d             Ô         Ĕ             #         $  S         T                                              C         D  s         t             ӡ         ԡ             3         4  c         d             ç         ħ             #         $  S         T                                              C         D  s                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     L                       <!  1          1  V=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     -#-    #-w5-    w5-D-    D-_w-l    _w-A4-    A4--x    -ub-	    ub-9-
    9-G-
    G--|    -ճ-    ճ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               =!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         +             V  K            v                                                          4	  
          S
  :         r                                   -         .  \           &         '  U         V                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |        (#  d$          T5  :6          u6  [7          7  |8          8  9          9  :         CF  G         G  H         H  I         I  9K         :K  wL         xL  M         M  N         N  1P         2P  oQ         pQ  R         R  S         S  )U         *U  gV         hV  W         W  X         X  !Z         "Z  _[         `[  \         \  ]         ]  _         _  W`         X`  a         a  b         b  d         d  Oe         Pe  f         f  g         g  	i         
i  Gj         Hj  k         k  l         l  n         n  ?o         @o  }p         ~p  q         q  r         r  7t         8t  uu         vu  v         v  w         w  /y         0y  mz         nz  {         {  |         |  '~         (~  e         f                                    ]         ^             ن         چ             U         V             ы         ҋ             M         N             ɐ         ʐ             D         E                                   :         ;  x         y                        2         3  p         q                        *         +  h         i                        "         #  `         a             ܪ         ݪ             X         Y             ԯ         կ             P         Q             ̴                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               <!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    v
         w
             d      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    v
         w
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            4     -,-`     ,--     --    -);-    );-I-    I-}-d    }--    --p    -ob-    ob-?-
    ?-M-
    M--t    -9׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               7!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             .         /  ]         ^                                  r  ^                                                
          I  &          d  A            \            w                                             3          4   b!         c!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                        .  !          \  O            }                     d#  $         $  %         %  "'         $'  b(         d(  )         )  *         *  ",         $,  b-         d-  .         .  /         /  "1         $1  b2         d2  3         3  4         4  "6         $6  b7         d7  8         8  9         9  ";         $;  b<         d<  =         =  >         >  "@         $@  bA         dA  B         B  C         C  "E         $E  bF         dF  G         G  H         H  "J         $J  cK         dK  L         L  M         M  #O         $O  cP         dP  Q         Q  R         R  #T         $T  cU         dU  V         V  W         W  #Y         $Y  cZ         dZ  [         [  \         \  #^         $^  c_         d_  `         `  a         a  #c         $c  cd         dd  e         e  f         f  #h         $h  ci         di  j         j  k         k  #m         $m  cn         dn  o         ds  t         t  u         u  "w         $w  bx         dx  y         y  z         z  "|         $|  b}         d}  ~         ~             "         $  b         d                        "         $  b         d                        "         $  b         d                        "         $  b         d                        "         $  b         d                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               /!         J!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    v
         w
             d      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     F
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-);-x    );-I-|    I-}-\    }-2-    2--h    -b-    b-!n-	    !n--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               D!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             .         /  ]         ^                                                                                  .            I  ?          t  j                                                                     3          4   b!         c!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ?         @                                    ?         @                        	          
  ?         @                                    ?         @                                    ?         @                                    ?         @                                    ?         @               !          l9  b:          :  ;          ;  <          <  =          >  ?         F  H         H  GI         I  J         J  L         L  QM         SM  N         N  O         O  Q         Q  QR         SR  S         S  T         T  V         V  QW         SW  X         X  Y         Y  [         [  Q\         S\  ]         ]  ^         ^  `         `  Qa         Sa  b         b  c         c  e         e  Qf         Sf  g         g  h         h  j         j  Qk         Sk  l         l  m         m  o         o  p         p  #r         $r  cs           א         ؐ             W         X             ו         ؕ             W         X             ך         ؚ             W         X             ן         ؟             W         X             פ         ؤ             W         X             ש         ة             W         X             ׮         خ             W         X                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               G!         Z!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     I
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -#-    #-);-    );-I-    I-}-d    }--    --p    -ob-    ob-?-
    ?-M-
    M-簳-t    簳-9׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ;!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             .         /  ]         ^                                  r  ]                                                	          I  '          d  B            ]            x                                             3          4   b!         c!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                 B  5	          p	  c
          
           d#  $         $  %         %  #'         $'  c(         d(  )         )  *         *  #,         $,  c-         d-  .         .  /         /  #1         $1  c2         d2  3         3  4         4  #6         $6  c7         d7  8         8  9         9  #;         $;  c<         d<  =         =  >         >  #@         $@  cA         dA  B         B  C         C  #E         $E  cF         dF  G         G  H         H  #J         $J  bK         dK  L         L  M         M  "O         $O  bP         dP  Q         Q  R         R  "T         $T  bU         dU  V         V  W         W  "Y         $Y  bZ         dZ  [         [  \         \  "^         $^  b_         d_  `         `  a         a  "c         $c  bd         dd  e         e  f         f  "h         $h  bi         di  j         j  k         k  "m         $m  bn         dn  o         ds  t         t  u         u  #w         $w  cx         dx  y         y  z         z  #|         $|  c}         d}  ~         ~             #         $  c         d                        #         $  c         d                        #         $  c         d                        #         $  c         d                        #         $  c         d                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               /!         J!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    v
         w
             d      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     E
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -o -X     o -ְ-     ְ-7ܰ-    7ܰ-?4-p    ?4-?-T    ?-eh-4    eh-2-    2--8    -b-    b-!n-	    !n--8
    -9׳-
                   =         b      J      7   U   7   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               D!         T!  R1                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 64 })  outputElements  5184    input0Elements  31104   input1Elements  24576   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     4                           #         b      J      7   M     operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       @             .         /  ]         ^                       h	  F
          
  a            |                                              *            U  J                      I         J  x         y                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     4                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  1728    input0Elements  1728    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       x            >         @  ~                                  >         @  ~                      	          
  >         @  ~                                  >         @  ~                                  >         @  ~                                  >         @  ~                                  >         @  ~             !          H?  >@          t@  jA          A  B          B  C         F  H         H  FI         I  J         J  L         L  RM         SM  N         N  O         O  Q         Q  RR         SR  S         S  T         T  V         V  RW         SW  X         X  Y         Y  [         [  R\         S\  ]         ]  ^         ^  `         `  Ra         Sa  b         b  c         c  e         e  Rf         Sf  g         g  h         h  j         j  Rk         Sk  l         l  m         m  o         o  p         p  "r         $r  bs           ֐         ؐ             V         X             ֕         ؕ             V         X             ֚         ؚ             V         X             ֟         ؟             V         X             ֤         ؤ             V         X             ֩         ة             V         X             ֮         خ             V         X                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               E!         Z!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     J
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    <     -,-p     ,--     -#-    #-m6-    m6-E-    E-Ux-t    Ux-8-     8-w@-X    w@--    -{b-p	    {b-?-x
    ?-M-
    M--    -ֳ-\    ֳ--<                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               9!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     6            Q  .          l  I                                    		          D	  H
          
                                    N         O  }         ~                        E         F  t         u                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        zI  J         J  K         K  2M         4M  pN         rN  O         O  P         P  *R         ,R  hS         jS  T         T  U          U  V          $W  'X          bX  eY          Y  Z          Z  [         \  X]         Z]  ^         ^  _         _  a         a  Pb         Rb  c         c  d         d  
f         f  Hg         Jg  h         h  i         i  k         k  Al         Bl  m         m  n         n  o         o  9q         :q  wr         xr  s         s  t         t  1v         2v  ow         pw  x         x  y         y  ){         *{  g|         h|  }         }  ~         ~  !         "  _         `             ۃ         ܃             W         X             ӈ         Ԉ             O         P             ˍ         ̍  	         
  G         H             Ò         Ē             ?         @  }         ~                        7         8  u         v                        /         0  m         n                        '         (  e         f                                    ]         ^             ٬         ڬ             U         V             ѱ         ұ             L         N             ȶ         ޶             Y                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               :!         L!  1         1  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    v
         w
             d                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    v
         w
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            0     -,-X     ,--     -;-    ;-}5-x    }5-D-|    D-ew-\    ew-o0-    o0--h    -b-    b-x-	    x--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               F!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         +  !          V  L            w                                                          4	  
          S
  9         r                                   -         .  \           &         '  U         V                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |             
          C  M                                             %  	'         
'  9(         :(  i)         j)  *         *  +         +  ,         ,  ).         *.  Y/         Z/  0         0  1         1  2         2  4         4  I5         J5  y6         z6  7         7  8         8  	:         
:  9;         :;  i<         j<  =         =  >         >  ?         ?  )A         *A  YB         ZB  C         C  D         D  E         E  G         G  IH         JH  yI         h  i         i  *k         ,k  Zl         \l  m         m  n         n  o         o  q         q  Jr         Lr  zs         |s  t         t  u         u  
w         w  :x         <x  jy         ly  z         z  {         {  |         |  *~         ,~  Z         \                                              J         L  z         |             ڈ         ܈  
           :         <  j         P  ~                      ސ                      >         @  n         p             Ζ         Ж              .         0  ^         `                                               N         P  ~                      ޣ                      >         @  n         p             Ω         Щ              .         0  ^         `                                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               J!         X!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 u
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    <     -,-p     ,--     -/-    /-m6-    m6-E-    E-Ux-t    Ux-8-     8-w@-X    w@--    -b-p	    b-?-x
    ?-M-
    M--    -ֳ-\    ֳ--<                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               =!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     6            Q  /          l  J                                    	          D	  G
          
                                    N         O  }         ~                        E         F  t         u                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        zI  J         J  K         K  3M         4M  qN         rN  O         O  P         P  +R         ,R  iS         jS  T         T  U         U  #W         $W  aX         bX  Y         Y  Z         Z  \          \  ]          Z]  ]^          ^  _          _  `          a  b         Rb  c         c  d         d  f         f  Ig         Jg  h         h  i         i  k         k  @l         Bl  ~m         m  n         n  o         o  8q         :q  vr         xr  s         s  t         t  0v         2v  nw         pw  x         x  y         y  ({         *{  f|         h|  }         }  ~         ~            "  ^         `             ڃ         ܃             V         X             ҈         Ԉ             N         P             ʍ         ̍           
  F         H                      Ē              >         @  |         ~                        6         8  t         v                        .         0  l         n                        &         (  d         f                                    \         ^             ج         ڬ             T         V             б         ұ             M         N             ɶ         ޶             X                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               <!         L!  1         1  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    v
         w
             d                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    v
         w
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            0     -,-X     ,--     -;-    ;-}5-x    }5-D-|    D-ew-\    ew-o0-    o0--h    -b-    b-x-	    x--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               F!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         +             V  K            v                                                          4	  
          S
  :         r                                   -         .  \           &         '  U         V                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         O  Y                        	          
  "          [  e         %  '         
'  8(         :(  h)         j)  *         *  +         +  ,         ,  (.         *.  X/         Z/  0         0  1         1  2         2  4         4  H5         J5  x6         z6  7         7  8         8  :         
:  8;         :;  h<         j<  =         =  >         >  ?         ?  (A         *A  XB         ZB  C         C  D         D  E         E  G         G  HH         JH  xI         h  i         i  +k         ,k  [l         \l  m         m  n         n  o         o  q         q  Kr         Lr  {s         |s  t         t  u         u  w         w  ;x         <x  ky         ly  z         z  {         {  |         |  +~         ,~  [         \                                              K         L  {         |             ۈ         ܈             ;         <  k         P                        ߐ                      ?         @  o         p             ϖ         Ж              /         0  _         `                                               O         P                        ߣ                      ?         @  o         p             ϩ         Щ              /         0  _         `                                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               F!         T!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 u
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-/;-x    /;-I-|    I-}-\    }-'9-    '9--h    -b-    b-n-	    n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               :!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             .         /  ]         ^                                  r  ^                                                
          I  &          d  A            \            w                                             3          4   b!         c!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            :         <  v         x                        *         ,  f         h             	         	             V         X                        
           F         H                                   6         8  r         t                        &         (  b         d                                     R!         T!  "         "  #         #  %         %  C&         D&  '         '  (         (  )         )  3+         4+  o,         p,  -         -  .         .  #0         $0  _1         `1  2         2  3         3  5         5  O6         P6  7         7  8         8  :         :  ?;         @;  {<         |<  =         =  >         >  /@         0@  kA         lA  B         B  C         C  E          E  [F         \F  G          'a  ,b          jb  oc          c  d          d  e          3f  8g         m  o         o  Op           ֕         ؕ             N         P             ƚ         Ț             >         @  z         |                        .         0  j         l                                    Z         \             Ҫ         Ԫ             J         L             ¯         į              :         <  v         x                        *         ,  f         h                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               =!         R!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     +
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -S-    S-;-x    ;-_J-|    _J-}-\    }-I-    I--h    -]b-    ]b-ӂ-	    ӂ--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               N!         \!  1         
2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 	         	  L
         M
  {                                 C         D  r         s                                                 _  Y                                                            .!  "          I"  &#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            1         2  c         d                                   +         ,  ]         ^  	         	  
         
             %         &  W         X                                               Q         R                                              K         L  }         ~                                   E          F   w!         x!  "         "  #          /  /          <0  1          \1  62          |2  V3          3  v4         G  H         H  I         I  -K         .K  _L         `L  M         M  N         N  O         O  'Q         (Q  YR         ZR  S         S  T         T  U         U  !W         "W  SX         TX  Y         Y  Z         Z  [         [  ]         ]  M^         N^  _         _  `         `  a         a  c         c  Gd         Hd  ye         ze  f         f  g         g  i         i  Aj         Bj  sk         tk  l         l  m         Pp  q         q  r         r  s         s  u         u  Hv         Jv  zw         |w  x         x  y         y  {         {  B|         D|  t}         v}  ~         ~             
           <         >  n         p             ҅         ԅ             6         8  h         j             ̋         ΋              0         2  b         d             Ƒ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               )!         8!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-/;-x    /;-I-|    I-}-\    }-'9-    '9--h    -b-    b-	n-	    	n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               <!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             .         /  ]         ^                                  r  ]                                                	          I  '          d  B            ]            x                                             3          4   b!         c!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ;         <  w         x                        +         ,  g         h             	         	             W         X                                   G         H                                   7         8  s         t                        '         (  c         d                                     S!         T!  "         "  #         #  %         %  B&         D&  ~'         '  (         (  )         )  2+         4+  n,         p,  -         -  .         .  "0         $0  ^1         `1  2         2  3         3  5         5  N6         P6  7         7  8         8  :         :  >;         @;  z<         |<  =         =  >         >  .@         0@  jA         lA  B         B  C         C  E          E  ZF         \F  G          vg  {h          h  i          i  k          ?k  Dl          l  m         m  o         o  Np           ו         ؕ             O         P             ǚ         Ț             ?         @  {         |                        /         0  k         l                                    [         \             Ӫ         Ԫ             K         L             ï         į              ;         <  w         x                        +         ,  g         h                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               A!         V!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     *
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -G-    G-;-x    ;-eJ-|    eJ-}-\    }-I-    I--h    -cb-    cb-ӂ-	    ӂ--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               J!         X!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 "	         #	  Q
         R
                                   H         I  w         x                                    %            d  ]                                                            3!  "          N"  ,#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            0         2  b         d                                   *         ,  \         ^  	         	  
         
             $         &  V         X                                               P         R                                              J         L  |         ~                                   D          F   v!         x!  "         "  #          4  5          5  6          6  7          8  8          <9  :         G  H         H  I         I  ,K         .K  ^L         `L  M         M  N         N  O         O  &Q         (Q  XR         ZR  S         S  T         T  U         U   W         "W  RX         TX  Y         Y  Z         Z  [         [  ]         ]  L^         N^  ~_         _  `         `  a         a  c         c  Fd         Hd  xe         ze  f         f  g         g  i         i  @j         Bj  rk         tk  l         l  m         Pp  q         q  r         r  s         s  u         u  Iv         Jv  {w         |w  x         x  y         y  {         {  C|         D|  u}         v}  ~         ~                        =         >  o         p             Ӆ         ԅ             7         8  i         j             ͋         ΋              1         2  c         d             Ǒ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               +!         <!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-6-x    6-1E-|    1E-x-\    x--    --h    -b-    b-n-	    n--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               8!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     6            Q  .          l  I                                    	          D	  E
          
                                    N         O  }         ~                        S         T                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            8         :  r         t                                  "  Z         \             	         	           
  B         D  |         ~                        *         ,  d         f                                   L         N                                   4         6  n         p                        !         !  V"         X"  #         #  $          :%  &          \&  @'          ~'  b(          (  )          )  *         H  J         J  WK         XK  L         L  M         M  O         O  ?P         @P  yQ         zQ  R         R  S         S  'U         (U  aV         bV  W         W  X         X  Z         Z  I[         J[  \         \  ]         ]  ^         ^  1`         2`  ka         la  b         b  c         c  e         e  Sf         Tf  g         g  h         h  j         j  ;k         <k  ul         vl  m         No  p         p  q         q  r         r  5t         6t  ou         pu  v         v  w         w  y         y  Wz         Xz  {         {  |         |  ~         ~  ?         @  y         z                        '         (  a         b             Շ         և             I         J                                   1         2  k         l             ߒ                              >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               8!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     +
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -;-    ;-Y;-x    Y;-I-|    I-A}-\    A}--    --h    -Qb-    Qb-ق-	    ق--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               B!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             .         /  ]         ^                                                                                  .            I  :          t  e                                                          '         (  V          W   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            2         4  f         h                                   6         8  j         l  	         	  
         
             :         <  n         p                        
           >         @  r         t                                   B         D  v         x                                   F         H  z          |   !         !  "         "  $         $  %          &  3'         4'  g(         h(  )         )  *         *  ,         ,  7-         8-  k.         l.  /         /  0         0  2         2  ;3         <3  o4         p4  5         5  6         6  8         8  ?9         @9  s:         t:  ;         ;  <         <  >         >  C?         D?  w@         x@  A         A  B         B  D         D  GE         HE  {F         |F  G         G  H          b  c          d   e          >e  (f          ff  Pg          g  xh         "o  Tp         Vp  q         q  r         r  s         s  $u         &u  Xv         Zv  w         w  x         x  y         y  ({         *{  \|         ^|  }         }  ~         ~             ,         .  `         b             Ȅ         ʄ             0         2  d         f             ̊         Ί              4         6  h         j             А         Ґ             8                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               &!         .!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-6-x    6-+E-|    +E-yx-\    yx--    --h    -b-    b-n-	    n--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               :!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     6            Q  /          l  J                                    	          D	  F
          
                                    N         O  }         ~                        Q         R                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            9         :  s         t                        !         "  [         \             	         	  	         
  C         D  }         ~                        +         ,  e         f                                   M         N                                   5         6  o         p                        !         !  W"         X"  #         #  $          *  +          ,  ,          (-  .          J.  ./          l/  P0         H  J         J  VK         XK  L         L  M         M  O         O  >P         @P  xQ         zQ  R         R  S         S  &U         (U  `V         bV  W         W  X         X  Z         Z  H[         J[  \         \  ]         ]  ^         ^  0`         2`  ja         la  b         b  c         c  e         e  Rf         Tf  g         g  h         h   j         j  :k         <k  tl         vl  m         No  p         p  q         q  r         r  4t         6t  nu         pu  v         v  w         w  y         y  Vz         Xz  {         {  |         |  ~         ~  >         @  x         z                        &         (  `         b             ԇ         և             H         J                                   0         2  j         l             ޒ                              >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               :!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     *
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-S;-x    S;-I-|    I-;}-\    ;}--    --h    -Wb-    Wb-ق-	    ق--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               >!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             .         /  ]         ^                                                                                  .            I  9          t  d                                                          "         #  Q          R   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            3         4  g         h                                   7         8  k         l  	         	  
         
             ;         <  o         p                                   ?         @  s         t                                   C         D  w         x                                   G         H  {          |   !         !  "         "  $         $  %          &  2'         4'  f(         h(  )         )  *         *  ,         ,  6-         8-  j.         l.  /         /  0         0  2         2  :3         <3  n4         p4  5         5  6         6  
8         8  >9         @9  r:         t:  ;         ;  <         <  >         >  B?         D?  v@         x@  A         A  B         B  D         D  FE         HE  zF         |F  G         G  H          h  i          i  j          k  k          .l  m          Vm  @n         "o  Up         Vp  q         q  r         r  s         s  %u         &u  Yv         Zv  w         w  x         x  y         y  ){         *{  ]|         ^|  }         }  ~         ~             -         .  a         b             Ʉ         ʄ             1         2  e         f             ͊         Ί             5         6  i         j             ѐ         Ґ             9                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               (!         2!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -G-    G-;;-x    ;;-I-|    I-#}-\    #}- -     --h    -b-    b-qx-	    qx--h
    -K׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               K!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      4         5  c	         d	  
         
           ]                                              G          H             n  F            l                                              1            \  =              h!          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            7         8  o         p                                   O         P             	         	  
         
  /         0  g         h                                   G         H                                   '         (  _         `                                   ?         @  w         x                          "          "  W#         X#  $         $  %         %  &          '  6(         8(  n)         p)  *         *  +         +  -         -  N.         P.  /         /  0         0  1         1  .3         03  f4         h4  5         5  6         6  8         8  F9         H9  ~:         :  ;         ;  <         <  &>         (>  ^?         `?  @         @  A         A  C         C  >D         @D  vE         xE  F         F  G         G  I         I  K         K  ML         NL  M         M  N         N  O         O  -Q         .Q  eR         fR  S         S  T         T  V         V  EW         FW  }X         ~X  Y         Y  Z         Z  %\         &\  ]]         ^]  ^         ^  _         _  a         a  =b         >b  uc         vc  d         d  e         e  g         g  Uh         Vh  i         i  j         j  k         k  5m         6m  mn          {  ~|          |  }          ~  ~          @  ,          z  f                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               D!         J!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 a
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -A-    A-;-x    ;-J-|    J-}-\    }--    --P    -b-    b-c-	    c-=-D
    =-K׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               F!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 6	         7	  e
         f
                        -         .  \         ]                                               9  *          x  i                                              ,   	!          G!  $"          b"  ?#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            5         6  k         l                                   C         D  y         z  	         	  
         
             Q         R                                   )         *  _         `                                   7         8  m         n                                   E         F  {         |               !         !  #         #  S$         T$  %         %  &         &  '         '  +)         ,)  a*         b*  +         +  ,         ,  .         .  9/         :/  o0         p0  1         1  2         2  4         4  G5         H5  }6         ~6  7         7  8         8  :          :  U;         V;  <         <  =         =  >         >  -@         .@  cA         dA  B         B  C         C  E         E  ;F         <F  pG         rG  H         H  I         nn  o         o  p         p  r         r  Ds         Fs  zt         |t  u         u  v         v  x         x  Ry         Ty  z         z  {         {  |         |  *~         ,~  `         b             ́         ΁             8         :  n         p             ڇ         ܇             F         H  |         ~                       :  ?          }              Ŗ                      F  K                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               >!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 e
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -A-    A-5;-x    5;-I-|    I-}-\    }--    --h    -b-    b-qx-	    qx--h
    -K׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               G!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      4         5  c	         d	  
         
           ]                                              G          H             m  E            j                                              ,            W  7              b!          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            6         8  n         p                                   N         P             	         	  
         
  .         0  f         h                                   F         H  ~                                 &         (  ^         `                                   >         @  v         x                          "          "  V#         X#  $         $  %         %  &          '  7(         8(  o)         p)  *         *  +         +  -         -  O.         P.  /         /  0         0  1         1  /3         03  g4         h4  5         5  6         6  8         8  G9         H9  :         :  ;         ;  <         <  '>         (>  _?         `?  @         @  A         A  C         C  ?D         @D  wE         xE  F         F  G         G  I         I  K         K  LL         NL  M         M  N         N  O         O  ,Q         .Q  dR         fR  S         S  T         T  V         V  DW         FW  |X         ~X  Y         Y  Z         Z  $\         &\  \]         ^]  ^         ^  _         _  a         a  <b         >b  tc         vc  d         d  e         e  g         g  Th         Vh  i         i  j         j  k         k  4m         6m  ln                        ڃ          (            b  N                               >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               F!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 a
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -;-    ;-;-x    ;-J-|    J-}-\    }-)-    )--P    -b-    b-c-	    c-=-D
    =-K׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               E!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 ;	         <	  j
         k
                        2         3  a         b                                               >  0          }  o                                              1   !          L!  *"          g"  E#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            4         6  j         l                                   B         D  x         z  	         	  
         
             P         R                                   (         *  ^         `                                    6         8  l         n                                   D         F  z         |               !         !  #         #  R$         T$  %         %  &         &  '         '  *)         ,)  `*         b*  +         +  ,         ,  .         .  8/         :/  n0         p0  1         1  2         2  4         4  F5         H5  |6         ~6  7         7  8         8  :          :  T;         V;  <         <  =         =  >         >  ,@         .@  bA         dA  B         B  C         C  E         E  :F         <F  qG         rG  H         H  I         nn  o         o  p         p  r         r  Es         Fs  {t         |t  u         u  v         v  x         x  Sy         Ty  z         z  {         {  |         |  +~         ,~  a         b             ́         ΁             9         :  o         p             ۇ         ܇             G         H  }         ~                                   ̚  ћ                      R  W                               >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               @!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 e
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -G-    G-y6-x    y6-E-|    E-ax-\    ax-7-    7--h    -{b-    {b-wx-	    wx--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               K!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     6            Q  .          l  I                                    	          D	  @
          
                                    N         O  }         ~                        E         F  t         u                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                      D  4          x  h                                 U$  %         %  &         &  '         '  ,)         -)  b*         c*  +         +  ,         ,  .         .  :/         ;/  p0         q0  1         1  2         2  4         4  H5         I5  ~6         6  7         7  8         8   :         !:  V;         W;  <         <  =         =  >         >  .@         /@  dA         eA  B         B  C         C  E         E  <F         =F  rG         G  I         I  CJ         EJ  yK         {K  L         L  M         M  O         O  QP         SP  Q         Q  R         R  S         S  )U         +U  _V         aV  W         W  X         X  Z         Z  7[         9[  m\         o\  ]         ]  ^         ^  `         `  Ea         Ga  {b         }b  c         c  d         d  f         f  Sg         Ug  h         h  i         i  j         j  +l         l  
n         ړ             E         F  {         |                                   S         T                                   +         ,  a         b             ͤ         Τ             9         :  o         p             ۪         ܪ             G         H  }         ~                                    U         V                      µ             -                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ;!         <!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 a
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     -/-    /-5;-    5;-I-    I-}-l    }--    ---P    ---    -b-h	    b-K-p
    K-Y-
    Y-۰-    ۰-E׳-L                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               A!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             .         /  ]         ^                                                                                  .            I  7          t  b                                                                     3          4   b!         c!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        T#  $         $  %         %  &         &  3(         4(  k)         l)  *         *  +         +  -         -  K.         L.  /         /  0         0  1         1  +3         ,3  c4         d4  5         5  6         6  8         8  C9         D9  {:         |:  ;         ;  <         <  #>         $>  [?         \?  @         @  A         A  C         C  ;D         <D  sE         tE  F         F  G         G  I         I  RJ         TJ  K         K  L         L  M         M  2O         4O  jP         lP  Q         Q  R         R  T          T  U          LU  IV          V  W          W  X          X  Y         ,Z  b[         d[  \         \  ]         ]  
_         _  B`         D`  za         |a  b         b  c         c  "e         $e  Zf         \f  g         g  h         h  j         j  :k         <k  rl         tl  m         m  n         n  p         p  Rq         Tq  r         r  s         s  t         t  2v         4v  jw         lw  x         x  y         y  {         {  J|         L|  }         }  ~         ~             *         ,  b         d             ҄         Ԅ  
           B         D  z         |                        "         $  Z         \             ʏ         ̏                                b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               I!         Z!  1         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    v
         w
             d                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     &
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -;-    ;-y6-x    y6-E-|    E-ax-\    ax-7-    7--h    -b-    b-qx-	    qx--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               G!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     6            Q  /          l  J                                     	          D	  ?
          
  ~                                  N         O  }         ~                        E         F  t         u                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                     H  8           |   l!          !  "          "  #         U$  %         %  &         &  '         '  +)         -)  a*         c*  +         +  ,         ,  .         .  9/         ;/  o0         q0  1         1  2         2  4         4  G5         I5  }6         6  7         7  8         8  :         !:  U;         W;  <         <  =         =  >         >  -@         /@  cA         eA  B         B  C         C  E         E  ;F         =F  qG         G  I         I  DJ         EJ  zK         {K  L         L  M         M  O         O  RP         SP  Q         Q  R         R  S         S  *U         +U  `V         aV  W         W  X         X  Z         Z  8[         9[  n\         o\  ]         ]  ^         ^  `         `  Fa         Ga  |b         }b  c         c  d         d  f         f  Tg         Ug  h         h  i         i  j         j  ,l         l  	n         ړ             D         F  z         |                                   R         T                                   *         ,  `         b             ̤         Τ             8         :  n         p             ڪ         ܪ             F         H  |         ~                                    T         V                      µ             ,                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               =!         @!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 a
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     -/-    /-5;-    5;-I-    I-}-l    }--    ---P    ---    -b-h	    b-K-p
    K-Y-
    Y-۰-    ۰-E׳-L                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               A!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             .         /  ]         ^                                                                                  .            I  6          t  a                                                                     3          4   b!         c!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        T#  $         $  %         %  &         &  2(         4(  j)         l)  *         *  +         +  -         -  J.         L.  /         /  0         0  1         1  *3         ,3  b4         d4  5         5  6         6  
8         8  B9         D9  z:         |:  ;         ;  <         <  ">         $>  Z?         \?  @         @  A         A  C         C  :D         <D  rE         tE  F         F  G         G  I         I  SJ         TJ  K         K  L         L  M         M  3O         4O  kP         lP  Q         Q  R         R  T         T  KU         LU  V         V  W         W  X         X  +Z          ,Z  )[          d[  a\          \  ]          ]  ^          _  	`         D`  {a         |a  b         b  c         c  #e         $e  [f         \f  g         g  h         h  j         j  ;k         <k  sl         tl  m         m  n         n  p         p  Sq         Tq  r         r  s         s  t         t  3v         4v  kw         lw  x         x  y         y  {         {  K|         L|  }         }  ~         ~             +         ,  c         d             ӄ         Ԅ             C         D  {         |                        #         $  [         \             ˏ         ̏                                b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               I!         Z!  1         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    v
         w
             d                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     %
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-g-     g--    -#;-x    #;-I-|    I-}-\    }-A-    A--h    -b-    b-	-	    	--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ,!         0!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      4         5  c	         d	  
         
           ]                                              G          H  ,          g  K            j                                                          9  -           d   X!          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            2         4  f         h                                   6         8  j         l  	         	  
         
             :         <  n         p                        
           >         @  r         t                                   B         D  v         x                                   F         H  z          |   !         !  "         "  $         $  J%         L%  ~&         &  '         '  (         (  *         *  N+         P+  ,         ,  -         -  .         .  0          0  R1         T1  2         2  3         3  4         4  "6         $6  V7         X7  8         8  9         9  :         :  &<         (<  Z=         \=  >         >  ?         ?  @         @  *B         ,B  ^C         `C  D         D  E         E  F         F  .H         0H  bI         dI  J         J  K         K  L          M  2N         4N  fO         hO  P         P  Q         Q  S         S  6T         8T  jU         lU  V         W  QX         SX  Y         Y  Z         Z  [         [  !]         #]  U^         W^  _         _  `         `  a         a  %c         'c  Yd         [d  e         e  f         f  g         g  )i         +i  ]j          _j  Mk          m  n          n  o          p  q          Cq  1r         T                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               8!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   "      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     -/-    /-5-    5-D-    D-kw-l    kw-U2-    U2--x    -b-	    b-K-
    K-Y-
    Y--|    -ֳ-    ֳ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ?!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         +            V  D            o                                                          4	  
          S
  1         r                                   -         .  \           &         '  U         V                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                      H  5          p  ]                                 "  #         #  !%         #%  [&         ]&  '         '  (         (  	*         *  C+         E+  },         ,  -         -  .         .  +0         -0  e1         g1  2         2  3         3  5         5  M6         O6  7         7  8         8  9         9  5;         7;  o<         q<  =         =  >         >  @         @  WA         YA  B         B  C         C  E         E  ?F         h  8i         :i  rj         tj  k         k  l         l   n         "n  Zo         \o  p         p  q         q  s         
s  Bt         Dt  |u         ~u  v         v  w         w  *y         ,y  dz         fz  {         {  |         |  ~         ~  L         N                                   4         6  n         p                                   V         X             ʌ         ̌             =         ?  w         y                        #         %  ]         _             ї         ӗ             E         G                                   -         /  g         i             ۢ         ݢ             O         Q             ç         ŧ             7         9  q         s                                 !  Y         [             Ͳ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               >!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    v
         w
             d      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    v
         w
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     q
            0     -,-X     ,-a-     a--    -#;-x    #;-I-|    I-}-\    }-A-    A--h    -ub-    ub-	-	    	--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               )!         .!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      4         5  c	         d	  
         
           ]                                              G          H  -          g  L            k                                                          9  ,           d   W!          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            3         4  g         h                                   7         8  k         l  	         	  
         
             ;         <  o         p                                   ?         @  s         t                                   C         D  w         x                                   G         H  {          |   !         !  "         "  $         $  K%         L%  &         &  '         '  (         (  *         *  O+         P+  ,         ,  -         -  .         .  0          0  S1         T1  2         2  3         3  4         4  #6         $6  W7         X7  8         8  9         9  :         :  '<         (<  [=         \=  >         >  ?         ?  @         @  +B         ,B  _C         `C  D         D  E         E  F         F  /H         0H  cI         dI  J         J  K         K  L          M  3N         4N  gO         hO  P         P  Q         Q  S         S  7T         8T  kU         lU  V         W  RX         SX  Y         Y  Z         Z  [         [  "]         #]  V^         W^  _         _  `         `  a         a  &c         'c  Zd         [d  e         e  f         f  g         g  *i         +i  ^j          ir  Ws          s  }t          t  u          u  v          w  w         T                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               4!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   !      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     -/-    /-5-    5-D-    D-kw-l    kw-U2-    U2--x    -b-	    b-K-
    K-Y-
    Y--|    -ֳ-    ֳ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ?!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         +            V  C            n                                                          4	  
          S
  2         r                                   -         .  \           &         '  U         V                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                 8  %           `   M!          !  u"         "  #         #  "%         #%  \&         ]&  '         '  (         (  
*         *  D+         E+  ~,         ,  -         -  .         .  ,0         -0  f1         g1  2         2  3         3  5         5  N6         O6  7         7  8         8  9         9  6;         7;  p<         q<  =         =  >         >  @         @  XA         YA  B         B  C         C  E         E  @F         h  9i         :i  sj         tj  k         k  l         l  !n         "n  [o         \o  p         p  q         q  	s         
s  Ct         Dt  }u         ~u  v         v  w         w  +y         ,y  ez         fz  {         {  |         |  ~         ~  M         N                                   5         6  o         p                                   W         X             ˌ         ̌             >         ?  x         y                        $         %  ^         _             җ         ӗ             F         G                                   .         /  h         i             ܢ         ݢ             P         Q             ħ         ŧ             8         9  r         s                                  !  Z         [             β                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               <!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    v
         w
             d      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    v
         w
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     r
            0     -,-X     ,--     -A-    A-/;-x    /;-I-|    I-}-\    }--    --h    -b-    b--	    --h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               K!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             .         /  ]         ^                                  r  f                                                          I  &          d  A            \            w                                             3          4   b!         c!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            0         2  b         d                                   *         ,  \         ^  	         	  
         
             $         &  V         X                                               P         R                                              J         L  |         ~                                   D          F   v!         x!  "         "  #          L/  10          l0  Q1          1  q2          2  3          3  4         F  G         G  &I         (I  XJ         ZJ  K         K  L         L  M         M   O         "O  RP         TP  Q         Q  R         R  S         S  U         U  LV         NV  ~W         W  X         X  Y         Y  [         [  F\         H\  x]         z]  ^         ^  _         _  a         a  @b         Bb  rc         tc  d         d  e         e  g         
g  :h         <h  li         ni  j         j  k         k  m         m  5n         6n  go         ho  p         p  q         q  r         r  /t         0t  au         bu  v         v  w         w  x         x  )z         *z  [{         \{  |         |  }         }  ~         ~  #         $  U         V                                              O         P                                              I         J  {                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ;!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   "      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-/;-x    /;-I-|    I-}-\    }--    --h    -b-    b--n-	    -n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               >!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             .         /  ]         ^                                                                                  .            I  7          t  b                                                                     3          4   b!         c!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ;         <  w         x                        +         ,  g         h             	         	             W         X                        
           F         H                                   6         8  r         t                        &         (  b         d                                     R!         T!  "         "  #         #  %          %  &          &  '          *(  )          Z)  L*          *  |+         :H  uI         vI  J         J  K         K  )M         *M  eN         fN  O         O  P         P  R         R  US         VS  T         T  U         U  	W         
W  EX         FX  Y         Y  Z         Z  [         [  5]         6]  q^         r^  _         _  `         `  %b         &b  ac         bc  d         d  e         e  g         g  Qh         Rh  i         i  j         j  l         l  Am         Bm  |n         ~n  o         o  p         p  0r         2r  ls         ns  t         t  u         u   w         "w  \x         ^x  y         y  z         z  |         |  L}         N}  ~         ~                         <         >  x         z                        ,         .  h         j                                   X         Z             Џ         ҏ             H                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ?!         R!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     I
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -G-    G-/;-x    /;-I-|    I-}-\    }--    --h    -b-    b--	    --h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               M!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             .         /  ]         ^                                  r  e                                                          I  '          d  B            ]            x                                             3          4   b!         c!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            1         2  c         d                                   +         ,  ]         ^  	         	  
         
             %         &  W         X                                               Q         R                                              K         L  }         ~                                   E          F   w!         x!  "         "  #          4  5          6  6          ,7  8          L8  19          l9  Q:         F  G         G  'I         (I  YJ         ZJ  K         K  L         L  M         M  !O         "O  SP         TP  Q         Q  R         R  S         S  U         U  MV         NV  W         W  X         X  Y         Y  [         [  G\         H\  y]         z]  ^         ^  _         _  a         a  Ab         Bb  sc         tc  d         d  e         e  	g         
g  ;h         <h  mi         ni  j         j  k         k  m         m  4n         6n  fo         ho  p         p  q         q  r         r  .t         0t  `u         bu  v         v  w         w  x         x  (z         *z  Z{         \{  |         |  }         }  ~         ~  "         $  T         V                                              N         P                                              H         J  z                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               9!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   !      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-/;-x    /;-I-|    I-}-\    }--    --h    -b-    b--n-	    -n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               @!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             .         /  ]         ^                                                                                  .            I  6          t  a                                                                     3          4   b!         c!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            :         <  v         x                        *         ,  f         h             	         	             V         X                                   G         H                                   7         8  s         t                        '         (  c         d                                     S!         T!  "         "  #         #  %          +  ,          ,  -          .  /          J/  <0          z0  l1         :H  tI         vI  J         J  K         K  (M         *M  dN         fN  O         O  P         P  R         R  TS         VS  T         T  U         U  W         
W  DX         FX  Y         Y  Z         Z  [         [  4]         6]  p^         r^  _         _  `         `  $b         &b  `c         bc  d         d  e         e  g         g  Ph         Rh  i         i  j         j  l         l  @m         Bm  }n         ~n  o         o  p         p  1r         2r  ms         ns  t         t  u         u  !w         "w  ]x         ^x  y         y  z         z  |         |  M}         N}  ~         ~                        =         >  y         z                        -         .  i         j                                   Y         Z             я         ҏ             I                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ?!         R!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     J
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -;-    ;-;-    ;-I-    I-}-d    }-
-    
-W-0    W--    -b-H	    b-c-P
    c-O-
    O-9׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               F!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      4         5  c	         d	  
         
           ]                                              G          H  ,          g  K            j                                                          9  -           d   X!          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q   w  operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       \            .         0  ^         `                                               N         P  ~	         	  
         
                        >         @  n         p                                    .         0  ^         `                                               N         P  ~                                              >!         @!  n"         %  &         &  ((         *(  X)         Z)  *         *  +         +  ,         ,  .         .  H/         J/  x0         z0  1         1  2         2  4         
4  85         :5  h6         j6  7         7  8         8  9         9  (;         *;  X<         Z<  =         =  >         >  ?         ?  A         A  HB         JB  xC         zC  D         D  E         E  G         
G  8H         h   j         j  0k         2k  `l         bl  m         m  n         n  o         o   q         "q  Pr         Rr  s         s  t         t  u         u  w         w  @x         Bx  py         ry  z         z  {         {   }         }  0~         2~  `         b                                                            N  B         }             ۈ         ݈             ;                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               L!         l!  2         "2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 q
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-}5-x    }5-D-|    D-ew-\    ew-$-    $--h    -b-    b-'n-	    'n--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               >!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         +            V  D            o                                                          4	  
          S
  1         r                                   -         .  \           &         '  U         V                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            =         >  {         |                        5         6  s         t             	         	  -         .  k         l                        %         &  c         d                                   [         \                                   S         T                                   K          L   !         !  "         "  $         $  C%         G  H         H  0J         2J  nK         pK  L         L  M         M  (O         *O  fP         hP  Q         Q  R         R   T         "T  ^U         `U  V         V  W         W  Y         Y  VZ         XZ  [         [  \         \  ^         ^  N_         P_  `         `  a         a  c         
c  Fd         Hd  e         e  f         f   h         h  >i         @i  |j         ~j  k         k  l         n  o         o  *q         ,q  hr         jr  s         s  t         t  "v         $v  `w         bw  x         x  y         y  {         {  X|         Z|  }         }  ~         ~             P         R             ̃         ΃  
           H         J             Ĉ         ƈ             @         B  ~                                 8         :  v         x                       4  6          t  v                                  4  6                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               >!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     I
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -u -`     u -ְ-     ְ-=ܰ-    =ܰ--4-x    -4-?-\    ?-Sh-<    Sh-
-    
--    --    -b- 	    b-c-(
    c-O-
    O-9׳-
                   =         b      J      7   U   7   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               G!         R!  T1                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 64 })  outputElements  5184    input0Elements  31104   input1Elements  24576   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     4                           #         b      J      7   M     operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       @                      	         
  8         9  g	                      A         B  p         q                                                                        G  :          r  e                                          b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     4                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  1728    input0Elements  1728    invocations     1                           3         b      J      7   Q   w  operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       \            /         0  _         `                                               O         P  	         	  
         
                        ?         @  o         p                                    /         0  _         `                                               O         P                                                ?!         @!  o"         %  &         &  )(         *(  Y)         Z)  *         *  +         +  ,         ,  .         .  I/         J/  y0         z0  1         1  2         2  	4         
4  95         :5  i6         j6  7         7  8         8  9         9  );         *;  Y<         Z<  =         =  >         >  ?         ?  A         A  IB         JB  yC         zC  D         D  E         E  	G         
G  9H         h  j         j  1k         2k  al         bl  m         m  n         n  o         o  !q         "q  Qr         Rr  s         s  t         t  u         u  w         w  Ax         Bx  qy         ry  z         z  {         {  }         }  1~         2~  a         b                                              M         N  |          }  r                      ݈  ҉                                          b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     4                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               J!         h!  1         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 q
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-}5-x    }5-D-|    D-ew-\    ew-*-    *--h    -b-    b-'n-	    'n--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               @!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         +            V  C            n                                                          4	  
          S
  2         r                                   -         .  \           &         '  U         V                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            <         >  z         |                        4         6  r         t             	         	  ,         .  j         l                        $         &  b         d                                   Z         \                                   R         T                                   J          L   !         !  "         "  $         $  B%         G  H         H  1J         2J  oK         pK  L         L  M         M  )O         *O  gP         hP  Q         Q  R         R  !T         "T  _U         `U  V         V  W         W  Y         Y  WZ         XZ  [         [  \         \  ^         ^  O_         P_  `         `  a         a  	c         
c  Gd         Hd  e         e  f         f  h         h  ?i         @i  }j         ~j  k         k  l         n  o         o  +q         ,q  ir         jr  s         s  t         t  #v         $v  aw         bw  x         x  y         y  {         {  Y|         Z|  }         }  ~         ~             Q         R             ̓         ΃             I         J             ň         ƈ             A         B                                   9         :  w         x                       t  v                                  4  6          t  v                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               <!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     J
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -A-    A-);-x    );-I-|    I-}-\    }-7-    7--h    -b-    b-n-	    n--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               H!         X!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             .         /  ]         ^                                  r  f                                                          I  &          d  A            \            w                                             3          4   b!         c!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            >         @  ~                                  >         @  ~                      	          
  >         @  ~                                  >         @  ~                                  >         @  ~                                  >         @  ~                                  >         @  ~             !         !  "          #  >$         @$  ~%          %  &          &  '           (  )          @)  B*          *  +         K  L         L  /N         0N  oO         pO  P         P  Q         Q  /S         0S  oT         pT  U         U  V         V  /X         0X  oY         pY  Z         Z  [         [  /]         0]  o^         p^  _         _  `         `  /b         0b  oc         pc  d         d  e         e  /g         0g  oh         ph  i         i  j         j  /l         0l  om         pm  n         n  o         o  /q         0q  or           2         4  r         t                        2         4  r         t                        2         4  r         t                        2         4  r         t                        2         4  r         t                        2         4  r         t                        2         4  r         t                        2                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               K!         b!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     +
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-;-x    ;-}J-|    }J-}-\    }--    --h    -Wb-    Wb-x-	    x--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               @!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 E	         F	  t
         u
                        <         =  k         l                                   	            H  J                                                           ;   !          V!  3"          q"  N#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ?         @                                    ?         @                        	          
  ?         @                                    ?         @                                    ?         @                                    ?         @                                    ?         @               !         !  "          #  ?$         @$  %         %  &         &  '          (  ?)         @)  *         *  +         +  ,          -  ?.         @.  /         /  0         0  1          2  ?3         @3  4         4  5         5  6          7  ?8         @8  9         9  :         :  ;          <  ?=         @=  >         >  ?         ?  @          A  ?B         @B  C         C  D         D  E          F  ?G         @G  H         H  I         I  J          K  >L         @L  ~M         M  N         N  O          P  >Q         @Q  ~R         R  S         S  T          U  >V         @V  ~W         W  X         X  Y          Z  >[         @[  ~\         \  ]         ]  ^          _  >`         @`  ~a         a  b         b  c          d  >e         @e  ~f         f  g         g  h          i  >j         @j  ~k         k  l         l  m          p  ^                      ȇ                                       ^         `                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               +!         :!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 u
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -A-    A-);-x    );-I-|    I-}-\    }-7-    7--h    -b-    b-n-	    n--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               H!         X!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             .         /  ]         ^                                  r  e                                                          I  '          d  B            ]            x                                             3          4   b!         c!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ?         @                                    ?         @                        	          
  ?         @                                    ?         @                                    ?         @                                    ?         @                                    ?         @               !         !  "          #  ?$         @$  %          +  ,           -  .          @.  B/          /  0          0  1         K  L         L  .N         0N  nO         pO  P         P  Q         Q  .S         0S  nT         pT  U         U  V         V  .X         0X  nY         pY  Z         Z  [         [  .]         0]  n^         p^  _         _  `         `  .b         0b  nc         pc  d         d  e         e  .g         0g  nh         ph  i         i  j         j  .l         0l  nm         pm  n         n  o         o  .q         0q  nr           3         4  s         t                        3         4  s         t                        3         4  s         t                        3         4  s         t                        3         4  s         t                        3         4  s         t                        3         4  s         t                        3                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               O!         f!  2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     *
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-;-x    ;-J-|    J-}-\    }--    --h    -]b-    ]b-x-	    x--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               >!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 J	         K	  y
         z
                        A         B  p         q                                               M  N                                  
            %             @   !          [!  9"          v"  T#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            >         @  ~                                  >         @  ~                      	          
  >         @  ~                                  >         @  ~                                  >         @  ~                                  >         @  ~                                  >         @  ~             !         !  "          #  >$         @$  ~%         %  &         &  '          (  >)         @)  ~*         *  +         +  ,          -  >.         @.  ~/         /  0         0  1          2  >3         @3  ~4         4  5         5  6          7  >8         @8  ~9         9  :         :  ;          <  >=         @=  ~>         >  ?         ?  @          A  >B         @B  ~C         C  D         D  E          F  >G         @G  ~H         H  I         I  J          K  ?L         @L  M         M  N         N  O          P  ?Q         @Q  R         R  S         S  T          U  ?V         @V  W         W  X         X  Y          Z  ?[         @[  \         \  ]         ]  ^          _  ?`         @`  a         a  b         b  c          d  ?e         @e  f         f  g         g  h          i  ?j         @j  k         k  l         l  m          L  :          x  f                      Ў                          _         `                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               -!         >!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 u
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-g6-x    g6-E-|    E-Ox-\    Ox--    --h    -b-    b-n-	    n--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               @!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     6            Q  .          l  I                                    	          D	  @
          
                                    N         O  }         ~                        E         F  t         u                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            <         >  z         |                        4         6  r         t             	         	  ,         .  j         l                        $         &  b         d                                   Z         \                                   R         T                                   J          L   !         !  "         "  $         $  B%          $1  2          T2  F3          3  v4          4  5          5  6         H  !J         "J  _K         `K  L         L  M         M  O         O  WP         XP  Q         Q  R         R  T         T  OU         PU  V         V  W         W  	Y         
Y  GZ         HZ  [         [  \         \  ^         ^  ?_         @_  }`         ~`  a         a  b         b  7d         8d  ue         ve  f         f  g         g  /i         0i  mj         nj  k         k  l         l  'n         o  q         q  Er         Fr  s         s  t         t  u          v  =w         >w  {x         |x  y         y  z         z  5|         6|  s}         t}  ~         ~             -         .  k         l                        %         &  c         d             ߉                      [         \             ׎         ؎             S         T             ϓ         Г                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               @!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     +
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-s-     s--    -#;-x    #;-I-|    I-}-\    }-{-    {--h    -?b-    ?b-ǂ-	    ǂ--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               2!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             .         /  ]         ^                                                                                  .            I  7          t  b                                                                     3          4   b!         c!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            .         0  ^         `                                               N         P  ~	         	  
         
                        >         @  n         p                                    .         0  ^         `                                               N         P  ~                                              >!         @!  n"         p"  #         D%  s&         t&  '         '  (         (  *         *  3+         4+  c,         d,  -         -  .         .  /         /  #1         $1  S2         T2  3         3  4         4  5         5  7         7  C8         D8  s9         t9  :         :  ;         ;  =         =  3>         4>  c?         d?  @         @  A         A  B         B  #D         $D  SE         TE  F         F  G         G  H          JL  8M          vM  dN          N  O          O  P          P  Q         ir  s         s  t         t  u         u  'w         )w  Wx         Yx  y         y  z         z  {         {  }         }  G~         I~  w         y             ׁ         ف           	  7         9  g         i             Ǉ         ɇ             '         )  W         Y                                              G         I  w         y             ה         ٔ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               "!         ,!  1         1  ~=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-g6-x    g6-E-|    E-Ox-\    Ox--    --h    -b-    b-m-	    m--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               @!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     6            Q  /          l  J                                     	          D	  ?
          
  ~                                  N         O  }         ~                        E         F  t         u                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            =         >  {         |                        5         6  s         t             	         	  -         .  k         l                        %         &  c         d                                   [         \                                   S         T                                   K          L   !         !  "         "  $         $  C%          7  8          D8  69          t9  f:          :  ;          ;  <         H   J         "J  ^K         `K  L         L  M         M  O         O  VP         XP  Q         Q  R         R  T         T  NU         PU  V         V  W         W  Y         
Y  FZ         HZ  [         [  \         \   ^         ^  >_         @_  |`         ~`  a         a  b         b  6d         8d  te         ve  f         f  g         g  .i         0i  lj         nj  k         k  l         l  &n         o  q         q  Dr         Fr  s         s  t         t  u          v  <w         >w  zx         |x  y         y  z         z  4|         6|  r}         t}  ~         ~             ,         .  j         l                        $         &  b         d             މ                      Z         \             ֎         ؎             R         T             Γ         Г                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               B!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     *
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-m-     m--    -#;-x    #;-I-|    I-}-\    }--    --h    -Eb-    Eb-ǂ-	    ǂ--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               0!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             .         /  ]         ^                                                                                  .            I  6          t  a                                                                     3          4   b!         c!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            /         0  _         `                                               O         P  	         	  
         
                        ?         @  o         p                                    /         0  _         `                                               O         P                                                ?!         @!  o"         p"  #         D%  r&         t&  '         '  (         (  *         *  2+         4+  b,         d,  -         -  .         .  /         /  "1         $1  R2         T2  3         3  4         4  5         5  7         7  B8         D8  r9         t9  :         :  ;         ;  =         =  2>         4>  b?         d?  @         @  A         A  B         B  "D         $D  RE         TE  F         F  G         G  H          &R  S          RS  @T          ~T  lU          U  V          V  W         ir  s         s  t         t  u         u  (w         )w  Xx         Yx  y         y  z         z  {         {  }         }  H~         I~  x         y             ؁         ف           	  8         9  h         i             ȇ         ɇ             (         )  X         Y                                              H         I  x         y             ؔ         ٔ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               $!         0!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-;-x    ;-I-|    I-|-\    |--    --h    -b-    b-_x-	    _x--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               @!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      4         5  c	         d	  
         
           ]                                              G          H  ,          g  K            j                                                          9  -           d   X!          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            :         <  v         x                        *         ,  f         h             	         	             V         X                        
           F         H                                   6         8  r         t                        &         (  b         d                                     R!         T!  "         "  #         #  %         %  B&         D&  ~'         '  (         (  )         )  2+         4+  n,         p,  -         -  .         .  "0         $0  ^1         `1  2         2  3         3  5         5  N6         P6  7         7  8         8  :         :  >;         @;  z<         |<  =         =  >         >  .@         0@  jA         lA  B         B  C         C  E          E  ZF         \F  G         G  H         H  J         J  WK         XK  L         L  M         M  O         O  GP         HP  Q         Q  R         R  S         S  7U         8U  sV         tV  W         W  X         X  'Z         (Z  c[         d[  \         \  ]         ]  _         _  S`         T`  a         a  b         b  d         d  Ce         De  f         f  g         g  h         h  3j         4j  ok         pk  l         l  m         m  #o            Ӈ                      :  +          f  W                               >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               U!         V!   2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 a
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-y-     y--    -;-x    ;-wJ-|    wJ-}-\    }--    --P    -ub-    ub-c-	    c-%-D
    %--׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ?!         @!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 E	         F	  t
         u
                        <         =  k         l                                   	            H  J                                                           ;   !          V!  3"          q"  N#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            1         2  c         d                                   +         ,  ]         ^  	         	  
         
             %         &  W         X                                               Q         R                                              K         L  }         ~                                   E          F   w!         x!  "         "  #         #  %         %  ?&         @&  q'         r'  (         (  )         )  +         +  9,         :,  k-         l-  .         .  /         /  1         1  32         42  e3         f3  4         4  5         5  6         6  -8         .8  _9         `9  :         :  ;         ;  <         <  '>         (>  Y?         Z?  @         @  A         A  B         B  !D         "D  SE         TE  F         F  G         G  H         H  J          tV  rW          W  X          X  Y          (Z  &[          d[  b\         ir  s         s  t         t  u         u  /w         1w  ax         cx  y         y  z         z  {         {  )}         +}  [~         ]~                                   #         %  U         W                                              O         Q                                              I                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               C!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 e
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-;-x    ;-I-|    I-|-\    |--    --h    -b-    b-Yx-	    Yx--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               >!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      4         5  c	         d	  
         
           ]                                              G          H  -          g  L            k                                                          9  ,           d   W!          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ;         <  w         x                        +         ,  g         h             	         	             W         X                                   G         H                                   7         8  s         t                        '         (  c         d                                     S!         T!  "         "  #         #  %         %  C&         D&  '         '  (         (  )         )  3+         4+  o,         p,  -         -  .         .  #0         $0  _1         `1  2         2  3         3  5         5  O6         P6  7         7  8         8  :         :  ?;         @;  {<         |<  =         =  >         >  /@         0@  kA         lA  B         B  C         C  E          E  [F         \F  G         G  H         H  J         J  VK         XK  L         L  M         M  
O         O  FP         HP  Q         Q  R         R  S         S  6U         8U  rV         tV  W         W  X         X  &Z         (Z  b[         d[  \         \  ]         ]  _         _  R`         T`  a         a  b         b  d         d  Be         De  ~f         f  g         g  h         h  2j         4j  nk         pk  l         l  m         m  "o                        ێ                      B  3          n  _                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               Y!         Z!  2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 a
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-s-     s--    -;-x    ;-}J-|    }J-}-\    }--    --P    -ub-    ub-c-	    c-%-D
    %--׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               <!         >!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 J	         K	  y
         z
                        A         B  p         q                                               M  N                                  
            %             @   !          [!  9"          v"  T#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            0         2  b         d                                   *         ,  \         ^  	         	  
         
             $         &  V         X                                               P         R                                              J         L  |         ~                                   D          F   v!         x!  "         "  #         #  %         %  >&         @&  p'         r'  (         (  )         )  +         +  8,         :,  j-         l-  .         .  /         /   1         1  22         42  d3         f3  4         4  5         5  6         6  ,8         .8  ^9         `9  :         :  ;         ;  <         <  &>         (>  X?         Z?  @         @  A         A  B         B   D         "D  RE         TE  F         F  G         G  H         H  J          \  ]          ]  ^          _  `          T`  Ra          a  b         ir  s         s  t         t  u         u  0w         1w  bx         cx  y         y  z         z  {         {  *}         +}  \~         ]~                                   $         %  V         W                                              P         Q                                              J                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               C!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 e
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -{,-X     {,-1-     1--    -a6-x    a6-D-|    D-Ix-\    Ix--    --h    -{b-    {b-m-	    m--h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     R                       H!  1          1  n=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     2            K  $          d  =          }  {                                  :	  8
          y
  w                                 J         K  {         |                        I         J  z         {                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            4         6  j         l                                   B         D  x         z  	         	  
         
             P         R                                   (         *  ^         `                                    6         8  l         n                                   D         F  z         |               !         !  #         #  R$          I  UJ         VJ  K         K  L         L  M         M  -O         .O  cP         dP  Q         Q  R         R  T         T  ;U         <U  qV         rV  W         W  X         X  Z         Z  I[         J[  \         \  ]         ]  ^         ^  !`         "`  Wa         Xa  b         b  c         c  d         d  /f         0f  eg         fg  h         h  i         i  k         k  =l         >l  sm         tm  n         n  o         o  q         q  Kr         Lr  s         s  t         t  u         u  #w         $w  Yx         Zx  y         y  z         z  {         {  1}         2}  g~         h~             Ӏ         Ԁ  	         
  ?         @  u         v                                   M         N                                   %         &  [         \             Ǒ          ȑ              Г                      4            X  <                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               K!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     +
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,-I-     I--    -;-x    ;-I-|    I-|-\    |--    --h    -Eb-    Eb--	    --h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     ]            !          R!  1          1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             0         1  a         b                                                                                  &            ?  1          h  Z                                                                                  L!         M!  }"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            6         8  n         p                                   N         P             	         	  
         
  .         0  f         h                                   F         H  ~                                 &         (  ^         `                                   >         @  v         x                          "          "  V#         X#  $         $  %         %  &          '  7(         8(  o)         p)  *         *  +         +  -         -  O.         P.  /         /  0         0  1         1  /3         03  g4         h4  5         5  6         6  8         8  G9         H9  :         :  ;         ;  <         <  '>         (>  _?         `?  @         @  A         A  C         C  ?D         @D  wE         xE  F         F  G         G  I           I  J          VJ  LK          K  L          L  M          M  N         o   q         q  8r         :r  ps         rs  t         t  u         u  w         w  Px         Rx  y         y  z         z  {         {  0}         2}  h~         j~             ؀         ڀ             H         J                                   (         *  `         b             Ћ         ҋ           
  @         B  x         z                                  "  X                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               1!         2!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -{,-X     {,-1-     1--    -a6-x    a6-D-|    D-Ix-\    Ix-Y-    Y--h    -ub-    ub-m-	    m--h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     S                       H!  1          1  o=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     2            K  %          d  >          }  |                                  :	  9
          y
  x                                 J         K  {         |                        I         J  z         {                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            5         6  k         l                                   C         D  y         z  	         	  
         
             Q         R                                   )         *  _         `                                   7         8  m         n                                   E         F  {         |               !         !  #         #  S$          I  TJ         VJ  K         K  L         L  M         M  ,O         .O  bP         dP  Q         Q  R         R  T         T  :U         <U  pV         rV  W         W  X         X  Z         Z  H[         J[  ~\         \  ]         ]  ^         ^   `         "`  Va         Xa  b         b  c         c  d         d  .f         0f  dg         fg  h         h  i         i  k         k  <l         >l  rm         tm  n         n  o         o  q         q  Jr         Lr  s         s  t         t  u         u  "w         $w  Xx         Zx  y         y  z         z  {         {  0}         2}  f~         h~             Ҁ         Ԁ           
  >         @  t         v                                   L         N                                   $         &  Z         \             Ƒ          |  `                      ę              ̛                               >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               G!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     *
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,-C-     C--    -;-x    ;-I-|    I-|-\    |--    --h    -Qb-    Qb--	    --h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     \            !          N!  1          1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             0         1  a         b                                                                                  &             ?  0          h  Y                                                                                  L!         M!  }"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            7         8  o         p                                   O         P             	         	  
         
  /         0  g         h                                   G         H                                   '         (  _         `                                   ?         @  w         x                          "          "  W#         X#  $         $  %         %  &          '  6(         8(  n)         p)  *         *  +         +  -         -  N.         P.  /         /  0         0  1         1  .3         03  f4         h4  5         5  6         6  8         8  F9         H9  ~:         :  ;         ;  <         <  &>         (>  ^?         `?  @         @  A         A  C         C  >D         @D  vE         xE  F         F  G         G  I          .O  $P          dP  ZQ          Q  R          R  S          T  T         o  q         q  9r         :r  qs         rs  t         t  u         u  w         w  Qx         Rx  y         y  z         z  {         {  1}         2}  i~         j~             ـ         ڀ             I         J                                   )         *  a         b             ы         ҋ  	         
  A         B  y         z                        !         "  Y                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               5!         6!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,-C-     C--    -;-x    ;-I-|    I-|-\    |--    --h    -cb-    cb-Mx-	    Mx--h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     V                       P!  1          1  ~=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      .         /  _	         `	  
         
           g                                   *         +  [          \  8          }  Y            z                                              *            S  ;           |   d!          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            2         4  f         h                                   6         8  j         l  	         	  
         
             :         <  n         p                        
           >         @  r         t                                   B         D  v         x                                   F         H  z          |   !         !  "         "  $         $  J%         L%  ~&         &  '         '  (         (  *         *  N+         P+  ,         ,  -         -  .         .  0          0  R1         T1  2         2  3         3  4         4  "6         $6  V7         X7  8         8  9         9  :         :  &<         (<  Z=         \=  >         >  ?         ?  @         @  *B         ,B  ^C         `C  D         D  E         E  F         F  .H         ^H  I         I  J         J  K         K  -M         .M  aN         bN  O         O  P         P  Q         Q  1S         2S  eT         fT  U         U  V         V  X         X  5Y         6Y  iZ         jZ  [         [  \         \  ^         ^  9_         :_  m`         n`  a         a  b         b  	d         
d  =e         >e  qf         rf  g         g  h         h  j         j  Ak         Bk  ul          h|  f}          }  ~          ~            1  /          t  r                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               <!         >!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 _
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,-C-     C--    -;-x    ;-qJ-|    qJ-}-\    }-y&-    y&--P    -b-    b-c-	    c--D
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     ]            !          L!  1          1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 A	         B	  r
         s
                        @         A  q         r                                               R  F                                              (             A   !          Z!  3"          s"  L#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            9         :  s         t                        !         "  [         \             	         	  	         
  C         D  }         ~                        +         ,  e         f                                   M         N                                   5         6  o         p                        !         !  W"         X"  #         #  $         $  &         &  ?'         @'  y(         z(  )         )  *         *  ',         (,  a-         b-  .         .  /         /  1         1  I2         J2  3         3  4         4  5         5  17         27  k8         l8  9         9  :         :  <         <  S=         T=  >         >  ?         ?  A         A  ;B         <B  uC         vC  D         D  E         E  #G         $G  \H         vl  m         m  n         n  "p         $p  \q         ^q  r         r  s         s  
u         u  Dv         Fv  ~w         w  x         x  y         y  ,{         .{  f|         h|  }         }  ~         ~             N         P                      ă             6         8  p         r                                    X         Z             ̎                      ;  >          ~              ĥ                               >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               F!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 c
                     
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,-=-     =--    -;-x    ;-I-|    I-|-\    |-1-    1--h    -ib-    ib-Mx-	    Mx--h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     U                       L!  1          1  w=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      .         /  _	         `	  
         
           g                                   *         +  [          \  9          }  Z            {                                              *            S  :           |   c!          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            3         4  g         h                                   7         8  k         l  	         	  
         
             ;         <  o         p                                   ?         @  s         t                                   C         D  w         x                                   G         H  {          |   !         !  "         "  $         $  K%         L%  &         &  '         '  (         (  *         *  O+         P+  ,         ,  -         -  .         .  0          0  S1         T1  2         2  3         3  4         4  #6         $6  W7         X7  8         8  9         9  :         :  '<         (<  [=         \=  >         >  ?         ?  @         @  +B         ,B  _C         `C  D         D  E         E  F         F  /H         ^H  I         I  J         J  K         K  ,M         .M  `N         bN  O         O  P         P  Q         Q  0S         2S  dT         fT  U         U  V         V   X         X  4Y         6Y  hZ         jZ  [         [  \         \  ^         ^  8_         :_  l`         n`  a         a  b         b  d         
d  <e         >e  pf         rf  g         g  h         h  j         j  @k         Bk  tl                                  =  ;            ~          Ç                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               >!         B!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 _
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,-C-     C--    -;-x    ;-wJ-|    wJ-}-\    }-,-    ,--P    -b-    b-c-	    c--D
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     \            !          J!  1          1  {=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 F	         G	  w
         x
                        E         F  v         w                                               W  L                                              -             F    !          _!  9"          x"  R#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            8         :  r         t                                  "  Z         \             	         	           
  B         D  |         ~                        *         ,  d         f                                   L         N                                   4         6  n         p                        !         !  V"         X"  #         #  $         $  &         &  >'         @'  x(         z(  )         )  *         *  &,         (,  `-         b-  .         .  /         /  1         1  H2         J2  3         3  4         4  5         5  07         27  j8         l8  9         9  :         :  <         <  R=         T=  >         >  ?         ?   A         A  :B         <B  tC         vC  D         D  E         E  "G         $G  ]H         vl  m         m  n         n  #p         $p  ]q         ^q  r         r  s         s  u         u  Ev         Fv  w         w  x         x  y         y  -{         .{  g|         h|  }         }  ~         ~             O         P             Ã         ă             7         8  q         r                                    Y         Z             ͎          G  J                      ͩ  Ъ                      S  V                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               H!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 c
                     
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,-g-     g--    -m6-x    m6-E-|    E-Ux-\    Ux--    -e-h    e-b-    b-Sx-	    Sx--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               -!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     2            K  $          d  =          }  {                                  :	  8
          y
  w                                 J         K  {         |                        I         J  z         {                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         <  >                                              H  J         $  %         %  '         '  8(         :(  j)         l)  *         *  +         +   -         -  2.         4.  d/         f/  0         0  1         1  2         2  ,4         .4  ^5         `5  6         6  7         7  8         8  &:         (:  X;         Z;  <         <  =         =  >         >   @         "@  RA         TA  B         B  C         C  D         D  F         F  LG         NG  ~H         H  I         I  K         K  BL         DL  tM         vM  N         N  O         O  
Q         Q  <R         >R  nS         pS  T         T  U         U  W         W  7X         8X  iY         jY  Z         Z  [         [  \          ]  1^         2^  c_         d_  `         `  a         a  b         b  +d         ,d  ]e         ^e  f         f  g         g  h         h  %j         &j  Wk         Xk  l         l  m         m  n         n  p          p  Qq         Rq  r         r  s         s  t         t  v         v  Kw         Lw  }x         ~x  y         y  z         z  |         |  E}         F}  w~         x~             ۀ         ܀             ?         @  q         r             Ն         ֆ             9         :  k         l             ό         Ќ             3         d                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     L                       0!  |1          1  D=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 _
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,-y-     y--    -;-    ;-I-    I-|-l    |--    --P    --    -Wb-h	    Wb-?-p
    ?-M-
    M--    --׳-L                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               9!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             0         1  a         b                                                                                  &            ?  1          h  Z                                                                                  L!         M!  }"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        T#  $         $  %         %  '         '  C(         D(  )         )  *         *  +         +  3-         4-  o.         p.  /         /  0         0  #2         $2  _3         `3  4         4  5         5  7         7  O8         P8  9         9  :         :  <         <  ?=         @=  {>         |>  ?         ?  @         @  /B         0B  kC         lC  D         D  E         E  G          G  [H         \H  I         J  K         K  8M         :M  tN         vN  O         O  P         P  (R         *R  dS         fS  T         T  U         U  W         W  TX         VX  Y         Y  Z         Z  \         
\  D]         F]  ^         ^  _         _  `         `  4b         6b  pc         rc  d         d  e         e  $g         &g  `h         bh  i         i  j         j  l         l  Pm         Rm  n         n  o         o  q         q  @r         Br  |s         ~s  t         t  u         u  0w         2w  lx         nx  y         y  z         z   |         "|  \}         ^}  ~         ~                        L         N             Ą         Ƅ              <          >  =          z  y                                  .  -         j                                   X         Z                                b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     ]            !          ^!  1          2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    x
         y
             j                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                      
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -? -X     ? -հ-     հ-ܰ-    ܰ-s0-p    s0-1<-T    1<-d-4    d--    -k-8    k-	b-    	b-Sx-	    Sx--8
    --׳-
                   =         b      J      7   U   7   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               +!         >!  .1                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 64 })  outputElements  5184    input0Elements  31104   input1Elements  24576   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     4                           #         b      J      7   M     operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       @                                     2            K  %          d  c                                  !   	         `	  
         
                        #            P         Q                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     4                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  1728    input0Elements  1728    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       x                         !         $  %         %  '         '  9(         :(  k)         l)  *         *  +         +  -         -  3.         4.  e/         f/  0         0  1         1  2         2  -4         .4  _5         `5  6         6  7         7  8         8  ':         (:  Y;         Z;  <         <  =         =  >         >  !@         "@  SA         TA  B         B  C         C  D         D  F         F  MG         NG  H         H  I         I  K         K  CL         DL  uM         vM  N         N  O         O  Q         Q  =R         >R  oS         pS  T         T  U         U  W         W  6X         8X  hY         jY  Z         Z  [         [  \          ]  0^         2^  b_         d_  `         `  a         a  b         b  *d         ,d  \e         ^e  f         f  g         g  h         h  $j         &j  Vk         Xk  l         l  m         m  n         n  p          p  Pq         Rq  r         r  s         s  t         t  v         v  Jw         Lw  |x         ~x  y         y  z         z  |         |  D}         F}  v~         x~             ڀ         ܀             >         @  p         r             Ԇ         ֆ             8         :  j         l             Ό         Ќ              2          4  6          w  y         d                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     M                       4!  1          1  K=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 _
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,-y-     y--    -;-    ;-I-    I-|-l    |--    -m-P    m--    -Wb-h	    Wb-?-p
    ?-M-
    M--    --׳-L                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               7!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             0         1  a         b                                                                                  &             ?  0          h  Y                                                                                  L!         M!  }"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        T#  $         $  %         %  '         '  B(         D(  ~)         )  *         *  +         +  2-         4-  n.         p.  /         /  0         0  "2         $2  ^3         `3  4         4  5         5  7         7  N8         P8  9         9  :         :  <         <  >=         @=  z>         |>  ?         ?  @         @  .B         0B  jC         lC  D         D  E         E  G          G  ZH         \H  I         J  K         K  9M         :M  uN         vN  O         O  P         P  )R         *R  eS         fS  T         T  U         U  W         W  UX         VX  Y         Y  Z         Z  	\         
\  E]         F]  ^         ^  _         _  `         `  5b         6b  qc         rc  d         d  e         e  %g         &g  ah         bh  i         i  j         j  l         l  Qm         Rm  n         n  o         o  q         q  Ar         Br  }s         ~s  t         t  u         u  1w         2w  mx         nx  y         y  z         z  !|         "|  ]}         ^}  ~         ~                        M         N             ń         Ƅ             =         >  y         z                        -         .  i          j  i                                              Z  Y                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     \            !          ^!  1          2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    x
         y
             j                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                      
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-a-     a--    -#;-x    #;-I-|    I-}-\    }-+-    +-}-P    }-b-    b-c-	    c-7-D
    7-3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               *!         <!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      .         /  _	         `	  
         
           g                                   *         +  [          \  8          }  Y            z                                              *            S  ;           |   d!          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            >         C                                   J         O                        
         
  V         [                                 $  b         g                        +         0  n         s                        7         <  z                                  C         H                          "         "  O#         T#  $         $  %         %  '         '  [(         `(  )         )  *         *  $,         ),  g-         l-  .         .  /         /  01         51  s2         x2  3         3  4         4  <6         A6  7         7  8         8  :         
:  H;         M;  <         <  =         =  ?         ?  T@         Y@  A         A  B         B  D          "D  E          EE  &F          hF  IG          G  lH          I  J         ng  h         h  i         i  2k         7k  ul         zl  m         m  n          o  >p         Cp  q         q  r         r  t         t  Ju         Ou  v         v  w         w  y         y  Sz         Xz  {         {  |         |  ~         ~  ]         b                        &         +  l         m                        2         3  t         u                        =         >           4  r         w                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     K                       @!  1          1  [=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 q
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     -#-    #-5-    5-/D-    /D-}w-l    }w-A4-    A4--x    -{b-	    {b-?-
    ?-M-
    M--|    -ճ-    ճ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ;!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         )            R  D          {  m                                                          0	  
          Q
  7         r                                   5         6  f           <         =  m         n                                                       b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |        (#  c$          e$  I%          %  j&          &  '          '  (          (  )         CF  ~G         G  H         H  I         I  8K         :K  vL         xL  M         M  N         N  0P         2P  nQ         pQ  R         R  S         S  (U         *U  fV         hV  W         W  X         X   Z         "Z  ^[         `[  \         \  ]         ]  _         _  V`         X`  a         a  b         b  d         d  Ne         Pe  f         f  g         g  i         
i  Fj         Hj  k         k  l         l   n         n  >o         @o  |p         ~p  q         q  r         r  6t         8t  tu         vu  v         v  w         w  .y         0y  lz         nz  {         {  |         |  &~         (~  d         f                                    \         ^             ؆         چ             T         V             Ћ         ҋ             L         N             Ȑ         ʐ             C         E                                   9         ;  w         y                        1         3  o         q                        )         +  g         i                        !         #  _         a             ۪         ݪ             W         Y             ӯ         կ             O         Q             ˴                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               <!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    x
         y
             j      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    x
         y
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            0     -,-X     ,-a-     a--    -;-x    ;-I-|    I-}-\    }-+-    +-}-P    }-b-    b-c-	    c-7-D
    7-3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               )!         :!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      .         /  _	         `	  
         
           g                                   *         +  [          \  9          }  Z            {                                              *            S  :           |   c!          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            ?         C                                   K         O                        
         
  W         [                                  $  c         g                        ,         0  o         s                        8         <  {                                 D         H                          "         "  P#         T#  $         $  %         %  '         '  \(         `(  )         )  *         *  %,         ),  h-         l-  .         .  /         /  11         51  t2         x2  3         3  4         4  =6         A6  7         7  8         8  :         
:  I;         M;  <         <  =         =  ?         ?  U@         Y@  A         A  B         B  D          K  K          &L  M          IM  *N          lN  MO          O  pP         ng  h         h  i         i  3k         7k  vl         zl  m         m  n          o  ?p         Cp  q         q  r         r  t         t  Ku         Ou  v         v  w         w  y         y  Tz         Xz  {         {  |         |  ~         ~  ^         b                        '         +  k         m                        1         3  s         u                        <         >           4  s         w                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     L                       @!  1          1  \=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 r
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     --    -5-    5-/D-    /D-}w-l    }w-A4-    A4--x    -ob-	    ob-?-
    ?-M-
    M--|    -ճ-    ճ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               9!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         )            R  C          {  l                                                          0	  
          Q
  8         r                                   5         6  f           <         =  m         n                                                       b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |        (#  d$          
*  *          ++  ,          L,  0-          m-  Q.          .  r/         CF  G         G  H         H  I         I  9K         :K  wL         xL  M         M  N         N  1P         2P  oQ         pQ  R         R  S         S  )U         *U  gV         hV  W         W  X         X  !Z         "Z  _[         `[  \         \  ]         ]  _         _  W`         X`  a         a  b         b  d         d  Oe         Pe  f         f  g         g  	i         
i  Gj         Hj  k         k  l         l  n         n  ?o         @o  }p         ~p  q         q  r         r  7t         8t  uu         vu  v         v  w         w  /y         0y  mz         nz  {         {  |         |  '~         (~  e         f                                    ]         ^             ن         چ             U         V             ы         ҋ             M         N             ɐ         ʐ             D         E                                   :         ;  x         y                        2         3  p         q                        *         +  h         i                        "         #  `         a             ܪ         ݪ             X         Y             ԯ         կ             P         Q             ̴                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               8!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    x
         y
             j      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    x
         y
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            4     -,-`     ,-y-     y--    -;-    ;-I-    I-|-d    |--    --p    -ub-    ub-E-
    E-S-
    S-簳-t    簳-9׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               3!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             0         1  a         b                                  r  Z                                                          ?            X  1          q  J            c            |                                              L!         M!  }"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                 (            V  G            u         d#  $         $  %         %  "'         $'  b(         d(  )         )  *         *  ",         $,  b-         d-  .         .  /         /  "1         $1  b2         d2  3         3  4         4  "6         $6  b7         d7  8         8  9         9  ";         $;  b<         d<  =         =  >         >  "@         $@  bA         dA  B         B  C         C  "E         $E  bF         dF  G         G  H         H  "J         $J  cK         dK  L         L  M         M  #O         $O  cP         dP  Q         Q  R         R  #T         $T  cU         dU  V         V  W         W  #Y         $Y  cZ         dZ  [         [  \         \  #^         $^  c_         d_  `         `  a         a  #c         $c  cd         dd  e         e  f         f  #h         $h  ci         di  j         j  k         k  #m         $m  cn         dn  o         ds  t         t  u         u  "w         $w  bx         dx  y         y  z         z  "|         $|  b}         d}  ~         ~             "         $  b         d                        "         $  b         d                        "         $  b         d                        "         $  b         d                        "         $  b         d                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               3!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    x
         y
             j      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     B
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-;-x    ;-I-|    I-|-\    |-2-    2--h    -b-    b-!n-	    !n--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               @!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             0         1  a         b                                                                                  &            ?  1          h  Z                                                                                  L!         M!  }"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ?         @                                    ?         @                        	          
  ?         @                                    ?         @                                    ?         @                                    ?         @                                    ?         @               !          -  .          .  /          0   1          81  ,2          d2  X3         F  H         H  GI         I  J         J  L         L  QM         SM  N         N  O         O  Q         Q  QR         SR  S         S  T         T  V         V  QW         SW  X         X  Y         Y  [         [  Q\         S\  ]         ]  ^         ^  `         `  Qa         Sa  b         b  c         c  e         e  Qf         Sf  g         g  h         h  j         j  Qk         Sk  l         l  m         m  o         o  p         p  #r         $r  cs           א         ؐ             W         X             ו         ؕ             W         X             ך         ؚ             W         X             ן         ؟             W         X             פ         ؤ             W         X             ש         ة             W         X             ׮         خ             W         X                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               C!         V!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     E
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     --    -;-    ;-I-    I-|-d    |--    --p    -ub-    ub-E-
    E-S-
    S-簳-t    簳-9׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               7!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             0         1  a         b                                  r  Y                                                          ?            X  2          q  K            d            }                                              L!         M!  }"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                             <  -          j  [         d#  $         $  %         %  #'         $'  c(         d(  )         )  *         *  #,         $,  c-         d-  .         .  /         /  #1         $1  c2         d2  3         3  4         4  #6         $6  c7         d7  8         8  9         9  #;         $;  c<         d<  =         =  >         >  #@         $@  cA         dA  B         B  C         C  #E         $E  cF         dF  G         G  H         H  #J         $J  bK         dK  L         L  M         M  "O         $O  bP         dP  Q         Q  R         R  "T         $T  bU         dU  V         V  W         W  "Y         $Y  bZ         dZ  [         [  \         \  "^         $^  b_         d_  `         `  a         a  "c         $c  bd         dd  e         e  f         f  "h         $h  bi         di  j         j  k         k  "m         $m  bn         dn  o         ds  t         t  u         u  #w         $w  cx         dx  y         y  z         z  #|         $|  c}         d}  ~         ~             #         $  c         d                        #         $  c         d                        #         $  c         d                        #         $  c         d                        #         $  c         d                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               3!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    x
         y
             j      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     A
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-;-x    ;-I-|    I-|-\    |-2-    2--h    -b-    b-!n-	    !n--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               @!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             0         1  a         b                                                                                  &             ?  0          h  Y                                                                                  L!         M!  }"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            >         @  ~                                  >         @  ~                      	          
  >         @  ~                                  >         @  ~                                  >         @  ~                                  >         @  ~                                  >         @  ~             !          3  4          4  5          5  6          7  8          @8  49         F  H         H  FI         I  J         J  L         L  RM         SM  N         N  O         O  Q         Q  RR         SR  S         S  T         T  V         V  RW         SW  X         X  Y         Y  [         [  R\         S\  ]         ]  ^         ^  `         `  Ra         Sa  b         b  c         c  e         e  Rf         Sf  g         g  h         h  j         j  Rk         Sk  l         l  m         m  o         o  p         p  "r         $r  bs           ֐         ؐ             V         X             ֕         ؕ             V         X             ֚         ؚ             V         X             ֟         ؟             V         X             ֤         ؤ             V         X             ֩         ة             V         X             ֮         خ             V         X                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               A!         V!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     F
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    <     -,-p     ,--     --    -y6-    y6-E-    E-ax-t    ax-8-     8-w@-X    w@--    -ub-p	    ub-K-x
    K-Y-
    Y--    -ֳ-\    ֳ--<                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               5!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     2            K  $          d  =          }  {                                  :	  8
          y
  w                                 J         K  {         |                        I         J  z         {                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        zI  J         J  K         K  2M         4M  pN         rN  O         O  P         P  *R         ,R  hS         jS  T         T  U         U  "W         $W  `X         bX  Y         Y  Z         Z  \         \  X]         Z]  ^         ^  _         _  a         a  Pb          Rb  Sc          c  d          d  e          f  g          Jg  Kh         h  i         i  k         k  Al         Bl  m         m  n         n  o         o  9q         :q  wr         xr  s         s  t         t  1v         2v  ow         pw  x         x  y         y  ){         *{  g|         h|  }         }  ~         ~  !         "  _         `             ۃ         ܃             W         X             ӈ         Ԉ             O         P             ˍ         ̍  	         
  G         H             Ò         Ē             ?         @  }         ~                        7         8  u         v                        /         0  m         n                        '         (  e         f                                    ]         ^             ٬         ڬ             U         V             ѱ         ұ             L         N             ȶ         ޶             Y                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               6!         H!  1         1  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    x
         y
             j                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    x
         y
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     ~
            0     -,-X     ,--     -5-    5-5-x    5-5D-|    5D-w-\    w-2-    2--h    -b-    b-!n-	    !n--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               B!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         )            R  D          {  m                                                          0	  
          Q
  7         r                                   5         6  f           <         =  m         n                                                       b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            B         C                                   N         O                        
         
  Z         [                        #         $  f         g                        /         0  r         s                        ;         <  ~                                 G         H                          "         "  S#         T#  $         $  %         %  '         '  [(         `(  )         )  *         *  $,         ),  g-         l-  .         .  /         /  01         51  s2         x2  3         3  4         4  <6         A6  7         7  8         8  :         
:  H;         M;  <         <  =         =  ?         ?  T@         Y@  A         A  B         B  D         "D  `E         eE  F         i  j         j  7l         <l  zm         m  n         n   p         p  Cq         Hq  r         r  s         s  u         u  Ov         Tv  w         w  x         x  z         z  [{         `{  |         |  }         }  $         )  g         l                        0         5  s         x                        <         A                      ǋ           
  H         M             ΐ         Ӑ             T         Y            L  P                      į  Ȱ                       <  @                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               F!         T!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     E
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    <     -,-p     ,--     -)-    )-y6-    y6-E-    E-ax-t    ax-8-     8-w@-X    w@--    -{b-p	    {b-K-x
    K-Y-
    Y--    -ֳ-\    ֳ--<                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               9!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     2            K  %          d  >          }  |                                  :	  9
          y
  x                                 J         K  {         |                        I         J  z         {                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        zI  J         J  K         K  3M         4M  qN         rN  O         O  P         P  +R         ,R  iS         jS  T         T  U         U  #W         $W  aX         bX  Y         Y  Z         Z  \         \  Y]         Z]  ^         ^  _         _  a         a  Qb         Rb  c         c  d         d  f         f  Ig         Jg  h          h  i         i  k         k  @l         Bl  ~m         m  n         n  o         o  8q         :q  vr         xr  s         s  t         t  0v         2v  nw         pw  x         x  y         y  ({         *{  f|         h|  }         }  ~         ~            "  ^         `             ڃ         ܃             V         X             ҈         Ԉ             N         P             ʍ         ̍           
  F         H                      Ē              >         @  |         ~                        6         8  t         v                        .         0  l         n                        &         (  d         f                                    \         ^             ج         ڬ             T         V             б          ұ  Ӳ                      N  O                     ޶             X                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               8!         H!  1         1  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    x
         y
             j                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    x
         y
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     }
            0     -,-X     ,--     -5-    5-5-x    5-5D-|    5D-w-\    w-8-    8--h    -b-    b-'n-	    'n--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               B!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         )            R  C          {  l                                                          0	  
          Q
  8         r                                   5         6  f           <         =  m         n                                                       b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            A         C                        
           M         O                        
         
  Y         [                        "         $  e         g                        .         0  q         s                        :         <  }                                 F         H                          "         "  R#         T#  $         $  %         %  '         '  \(         `(  )         )  *         *  %,         ),  h-         l-  .         .  /         /  11         51  t2         x2  3         3  4         4  =6         A6  7         7  8         8  :         
:  I;         M;  <         <  =         =  ?         ?  U@         Y@  A         A  B         B  D         "D  aE         eE  F         i  j         j  8l         <l  {m         m  n         n  p         p  Dq         Hq  r         r  s         s  u         u  Pv         Tv  w         w  x         x  z         z  \{         `{  |         |  }         }  %         )  h         l                        1         5  t         x                        =         A             Ë         ǋ           
  I         M             ϐ         Ӑ             U         Y            x  |                                  ,  0          h  l                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               B!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     F
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-;-x    ;-I-|    I-}-\    }-'9-    '9--h    -b-    b-n-	    n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               6!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             0         1  a         b                                  r  Z                                                          ?            X  1          q  J            c            |                                              L!         M!  }"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            :         <  v         x                        *         ,  f         h             	         	             V         X                        
           F         H                                   6         8  r         t                        &         (  b         d                                     R!         T!  "         "  #         #  %         %  C&         D&  '         '  (         (  )         )  3+         4+  o,         p,  -         -  .         .  #0         $0  _1         `1  2         2  3         3  5         5  O6         P6  7         7  8         8  :         :  ?;         @;  {<         |<  =         =  >         >  /@         0@  kA         lA  B         B  C         C  E          E  [F         \F  G          G  H          H  I          I  J          .K   L          `L  RM         m  o         o  Op           ֕         ؕ             N         P             ƚ         Ț             >         @  z         |                        .         0  j         l                                    Z         \             Ҫ         Ԫ             J         L             ¯         į              :         <  v         x                        *         ,  f         h                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               9!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     +
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -G-    G-;-x    ;-J-|    J-}-\    }-I-    I--h    -Wb-    Wb-͂-	    ͂--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               J!         X!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 A	         B	  r
         s
                        @         A  q         r                                               R  F                                              (             A   !          Z!  3"          s"  L#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            1         2  c         d                                   +         ,  ]         ^  	         	  
         
             %         &  W         X                                               Q         R                                              K         L  }         ~                                   E          F   w!         x!  "         "  #          #  $          $  %          &  &          <'  (          \(  4)         G  H         H  I         I  -K         .K  _L         `L  M         M  N         N  O         O  'Q         (Q  YR         ZR  S         S  T         T  U         U  !W         "W  SX         TX  Y         Y  Z         Z  [         [  ]         ]  M^         N^  _         _  `         `  a         a  c         c  Gd         Hd  ye         ze  f         f  g         g  i         i  Aj         Bj  sk         tk  l         l  m         Pp  q         q  r         r  s         s  u         u  Hv         Jv  zw         |w  x         x  y         y  {         {  B|         D|  t}         v}  ~         ~             
           <         >  n         p             ҅         ԅ             6         8  h         j             ̋         ΋              0         2  b         d             Ƒ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               '!         6!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-;-x    ;-I-|    I-}-\    }-'9-    '9--h    -b-    b-n-	    n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               8!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             0         1  a         b                                  r  Y                                                          ?            X  2          q  K            d            }                                              L!         M!  }"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ;         <  w         x                        +         ,  g         h             	         	             W         X                                   G         H                                   7         8  s         t                        '         (  c         d                                     S!         T!  "         "  #         #  %         %  B&         D&  ~'         '  (         (  )         )  2+         4+  n,         p,  -         -  .         .  "0         $0  ^1         `1  2         2  3         3  5         5  N6         P6  7         7  8         8  :         :  >;         @;  z<         |<  =         =  >         >  .@         0@  jA         lA  B         B  C         C  E          E  ZF         \F  G          M  N          N  O          O  P          (Q  R          ZR  LS         m  o         o  Np           ו         ؕ             O         P             ǚ         Ț             ?         @  {         |                        /         0  k         l                                    [         \             Ӫ         Ԫ             K         L             ï         į              ;         <  w         x                        +         ,  g         h                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               =!         R!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     *
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -A-    A-;-x    ;-J-|    J-}-\    }-I-    I--h    -]b-    ]b-͂-	    ͂--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               F!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 F	         G	  w
         x
                        E         F  v         w                                               W  L                                              -             F    !          _!  9"          x"  R#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            0         2  b         d                                   *         ,  \         ^  	         	  
         
             $         &  V         X                                               P         R                                              J         L  |         ~                                   D          F   v!         x!  "         "  #          |)  T*          *  t+          +  ,          ,  -          -  .         G  H         H  I         I  ,K         .K  ^L         `L  M         M  N         N  O         O  &Q         (Q  XR         ZR  S         S  T         T  U         U   W         "W  RX         TX  Y         Y  Z         Z  [         [  ]         ]  L^         N^  ~_         _  `         `  a         a  c         c  Fd         Hd  xe         ze  f         f  g         g  i         i  @j         Bj  rk         tk  l         l  m         Pp  q         q  r         r  s         s  u         u  Iv         Jv  {w         |w  x         x  y         y  {         {  C|         D|  u}         v}  ~         ~                        =         >  o         p             Ӆ         ԅ             7         8  i         j             ͋         ΋              1         2  c         d             Ǒ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               )!         :!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-6-x    6-E-|    E-mx-\    mx--    --h    -{b-    {b-n-	    n--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               4!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     2            K  $          d  =          }                                    :	  =
          y
  |                                 J         K  {         |                        K         L  }         ~                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            8         :  r         t                                  "  Z         \             	         	           
  B         D  |         ~                        *         ,  d         f                                   L         N                                   4         6  n         p                        !         !  V"         X"  #         #  $          0  p1          1  2          2  3          3  4          5  5         H  J         J  WK         XK  L         L  M         M  O         O  ?P         @P  yQ         zQ  R         R  S         S  'U         (U  aV         bV  W         W  X         X  Z         Z  I[         J[  \         \  ]         ]  ^         ^  1`         2`  ka         la  b         b  c         c  e         e  Sf         Tf  g         g  h         h  j         j  ;k         <k  ul         vl  m         No  p         p  q         q  r         r  5t         6t  ou         pu  v         v  w         w  y         y  Wz         Xz  {         {  |         |  ~         ~  ?         @  y         z                        '         (  a         b             Շ         և             I         J                                   1         2  k         l             ߒ                              >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               4!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     +
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-A;-x    A;-I-|    I-)}-\    )}--    --h    -Kb-    Kb-ӂ-	    ӂ--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               >!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             0         1  a         b                                                                                  &            ?  ,          h  U            ~                                                         >          ?   o!         p!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            2         4  f         h                                   6         8  j         l  	         	  
         
             :         <  n         p                        
           >         @  r         t                                   B         D  v         x                                   F         H  z          |   !         !  "         "  $         $  %          &  3'         4'  g(         h(  )         )  *         *  ,         ,  7-         8-  k.         l.  /         /  0         0  2         2  ;3         <3  o4         p4  5         5  6         6  8         8  ?9         @9  s:         t:  ;         ;  <         <  >         >  C?         D?  w@         x@  A         A  B         B  D         D  GE         HE  {F         |F  G         G  H          ^W  FX          X  nY          Y  Z          Z  [          [  \         "o  Tp         Vp  q         q  r         r  s         s  $u         &u  Xv         Zv  w         w  x         x  y         y  ({         *{  \|         ^|  }         }  ~         ~             ,         .  `         b             Ȅ         ʄ             0         2  d         f             ̊         Ί              4         6  h         j             А         Ґ             8                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               "!         *!  1         1  |=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-6-x    6-%E-|    %E-sx-\    sx--    --h    -{b-    {b-n-	    n--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               6!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     2            K  %          d  >          }                                    :	  <
          y
  {                                 J         K  {         |                        M         N                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            9         :  s         t                        !         "  [         \             	         	  	         
  C         D  }         ~                        +         ,  e         f                                   M         N                                   5         6  o         p                        !         !  W"         X"  #         #  $          86  7          Z7  <8          |8  ^9          9  :          :  ;         H  J         J  VK         XK  L         L  M         M  O         O  >P         @P  xQ         zQ  R         R  S         S  &U         (U  `V         bV  W         W  X         X  Z         Z  H[         J[  \         \  ]         ]  ^         ^  0`         2`  ja         la  b         b  c         c  e         e  Rf         Tf  g         g  h         h   j         j  :k         <k  tl         vl  m         No  p         p  q         q  r         r  4t         6t  nu         pu  v         v  w         w  y         y  Vz         Xz  {         {  |         |  ~         ~  >         @  x         z                        &         (  `         b             ԇ         և             H         J                                   0         2  j         l             ޒ                              >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               6!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     *
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-A;-x    A;-I-|    I-)}-\    )}--    --h    -Qb-    Qb-ӂ-	    ӂ--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               :!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             0         1  a         b                                                                                  &             ?  +          h  T            }                                                       	  9          :   j!         k!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            3         4  g         h                                   7         8  k         l  	         	  
         
             ;         <  o         p                                   ?         @  s         t                                   C         D  w         x                                   G         H  {          |   !         !  "         "  $         $  %          &  2'         4'  f(         h(  )         )  *         *  ,         ,  6-         8-  j.         l.  /         /  0         0  2         2  :3         <3  n4         p4  5         5  6         6  
8         8  >9         @9  r:         t:  ;         ;  <         <  >         >  B?         D?  v@         x@  A         A  B         B  D         D  FE         HE  zF         |F  G         G  H          &]  ^          N^  6_          v_  ^`          `  a          a  b         "o  Up         Vp  q         q  r         r  s         s  %u         &u  Yv         Zv  w         w  x         x  y         y  ){         *{  ]|         ^|  }         }  ~         ~             -         .  a         b             Ʉ         ʄ             1         2  e         f             ͊         Ί             5         6  i         j             ѐ         Ґ             9                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               $!         .!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -A-    A-S;-x    S;-I-|    I-;}-\    ;}-5-    5--h    -b-    b-kx-	    kx--h
    -K׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               G!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      .         /  _	         `	  
         
           g                                   *         +  [          \  4            \                                              $            M  *          v  S              |!          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            7         8  o         p                                   O         P             	         	  
         
  /         0  g         h                                   G         H                                   '         (  _         `                                   ?         @  w         x                          "          "  W#         X#  $         $  %         %  &          '  6(         8(  n)         p)  *         *  +         +  -         -  N.         P.  /         /  0         0  1         1  .3         03  f4         h4  5         5  6         6  8         8  F9         H9  ~:         :  ;         ;  <         <  &>         (>  ^?         `?  @         @  A         A  C         C  >D         @D  vE         xE  F         F  G         G  I         I  K         K  ML         NL  M         M  N         N  O         O  -Q         .Q  eR         fR  S         S  T         T  V         V  EW         FW  }X         ~X  Y         Y  Z         Z  %\         &\  ]]         ^]  ^         ^  _         _  a         a  =b         >b  uc         vc  d         d  e         e  g         g  Uh         Vh  i         i  j         j  k         k  5m         6m  mn          և                        J  4            n                               >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               @!         F!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 _
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -;-    ;-<-x    <-J-|    J-}-\    }-w-    w--P    -b-    b-c-	    c-=-D
    =-K׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               B!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                  P	         Q	  
         
                        O         P                                              "            a  T                                              7             P   )!          i!  B"          "  [#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            5         6  k         l                                   C         D  y         z  	         	  
         
             Q         R                                   )         *  _         `                                   7         8  m         n                                   E         F  {         |               !         !  #         #  S$         T$  %         %  &         &  '         '  +)         ,)  a*         b*  +         +  ,         ,  .         .  9/         :/  o0         p0  1         1  2         2  4         4  G5         H5  }6         ~6  7         7  8         8  :          :  U;         V;  <         <  =         =  >         >  -@         .@  cA         dA  B         B  C         C  E         E  ;F         <F  pG         rG  H         H  I          >b  6c          vc  nd          d  e          e  f          g  h         nn  o         o  p         p  r         r  Ds         Fs  zt         |t  u         u  v         v  x         x  Ry         Ty  z         z  {         {  |         |  *~         ,~  `         b             ́         ΁             8         :  n         p             ڇ         ܇             F         H  |         ~                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               :!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 c
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -;-    ;-M;-x    M;-I-|    I-5}-\    5}-Y-    Y--h    -b-    b-kx-	    kx--h
    -K׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               C!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      .         /  _	         `	  
         
           g                                   *         +  [          \  4            [                                                          H  $          q  M              v!          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            6         8  n         p                                   N         P             	         	  
         
  .         0  f         h                                   F         H  ~                                 &         (  ^         `                                   >         @  v         x                          "          "  V#         X#  $         $  %         %  &          '  7(         8(  o)         p)  *         *  +         +  -         -  O.         P.  /         /  0         0  1         1  /3         03  g4         h4  5         5  6         6  8         8  G9         H9  :         :  ;         ;  <         <  '>         (>  _?         `?  @         @  A         A  C         C  ?D         @D  wE         xE  F         F  G         G  I         I  K         K  LL         NL  M         M  N         N  O         O  ,Q         .Q  dR         fR  S         S  T         T  V         V  DW         FW  |X         ~X  Y         Y  Z         Z  $\         &\  \]         ^]  ^         ^  _         _  a         a  <b         >b  tc         vc  d         d  e         e  g         g  Th         Vh  i         i  j         j  k         k  4m         6m  ln                      2            l  V                        ʓ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               B!         J!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 _
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-<-x    <-J-|    J-}-\    }-w-    w--P    -b-    b-c-	    c-=-D
    =-K׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               A!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 K	         L	  |
         }
                        J         K  {         |                                               \  N                                              2             K   %!          d!  >"          }"  W#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            4         6  j         l                                   B         D  x         z  	         	  
         
             P         R                                   (         *  ^         `                                    6         8  l         n                                   D         F  z         |               !         !  #         #  R$         T$  %         %  &         &  '         '  *)         ,)  `*         b*  +         +  ,         ,  .         .  8/         :/  n0         p0  1         1  2         2  4         4  F5         H5  |6         ~6  7         7  8         8  :          :  T;         V;  <         <  =         =  >         >  ,@         .@  bA         dA  B         B  C         C  E         E  :F         <F  qG         rG  H         H  I          Vh  Ni          i  j          j  k          k  l          6m  .n         nn  o         o  p         p  r         r  Es         Fs  {t         |t  u         u  v         v  x         x  Sy         Ty  z         z  {         {  |         |  +~         ,~  a         b             ́         ΁             9         :  o         p             ۇ         ܇             G         H  }         ~                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               <!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 c
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -A-    A-6-x    6-E-|    E-mx-\    mx-7-    7--h    -ub-    ub-qx-	    qx--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               G!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     2            K  $          d  =          }  s                                  :	  0
          y
  o                                 J         K  {         |                        I         J  z         {                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                        C  =                                             U$  %         %  &         &  '         '  ,)         -)  b*         c*  +         +  ,         ,  .         .  :/         ;/  p0         q0  1         1  2         2  4         4  H5         I5  ~6         6  7         7  8         8   :         !:  V;         W;  <         <  =         =  >         >  .@         /@  dA         eA  B         B  C         C  E         E  <F         =F  rG         G  I         I  CJ         EJ  yK         {K  L         L  M         M  O         O  QP         SP  Q         Q  R         R  S         S  )U         +U  _V         aV  W         W  X         X  Z         Z  7[         9[  m\         o\  ]         ]  ^         ^  `         `  Ea         Ga  {b         }b  c         c  d         d  f         f  Sg         Ug  h         h  i         i  j         j  +l         l  
n         ړ             E         F  {         |                                   S         T                                   +         ,  a         b             ͤ         Τ             9         :  o         p             ۪         ܪ             G         H  }         ~                                    U         V                      µ             -                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               7!         8!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 _
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     -)-    )-#;-    #;-I-    I-}-l    }--    ---P    ---    -b-h	    b-W-p
    W-e-
    e-۰-    ۰-E׳-L                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               =!         >!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             0         1  a         b                                                                                  &            ?  )          h  R            {                                                                      L!         M!  }"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        T#  $         $  %         %  &         &  3(         4(  k)         l)  *         *  +         +  -         -  K.         L.  /         /  0         0  1         1  +3         ,3  c4         d4  5         5  6         6  8         8  C9         D9  {:         |:  ;         ;  <         <  #>         $>  [?         \?  @         @  A         A  C         C  ;D         <D  sE         tE  F         F  G          G  H          I  J          TJ  OK          K  L          L  M         M  2O         4O  jP         lP  Q         Q  R         R  T         T  JU         LU  V         V  W         W  X         X  *Z         ,Z  b[         d[  \         \  ]         ]  
_         _  B`         D`  za         |a  b         b  c         c  "e         $e  Zf         \f  g         g  h         h  j         j  :k         <k  rl         tl  m         m  n         n  p         p  Rq         Tq  r         r  s         s  t         t  2v         4v  jw         lw  x         x  y         y  {         {  J|         L|  }         }  ~         ~             *         ,  b         d             ҄         Ԅ  
           B         D  z         |                        "         $  Z         \             ʏ         ̏                                b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               E!         V!  1         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    x
         y
             j                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     &
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-6-x    6-E-|    E-mx-\    mx-7-    7--h    -{b-    {b-qx-	    qx--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               C!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     2            K  %          d  >          }  t                                  :	  1
          y
  p                                 J         K  {         |                        I         J  z         {                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         O  I                        	          
            [  U         U$  %         %  &         &  '         '  +)         -)  a*         c*  +         +  ,         ,  .         .  9/         ;/  o0         q0  1         1  2         2  4         4  G5         I5  }6         6  7         7  8         8  :         !:  U;         W;  <         <  =         =  >         >  -@         /@  cA         eA  B         B  C         C  E         E  ;F         =F  qG         G  I         I  DJ         EJ  zK         {K  L         L  M         M  O         O  RP         SP  Q         Q  R         R  S         S  *U         +U  `V         aV  W         W  X         X  Z         Z  8[         9[  n\         o\  ]         ]  ^         ^  `         `  Fa         Ga  |b         }b  c         c  d         d  f         f  Tg         Ug  h         h  i         i  j         j  ,l         l  	n         ړ             D         F  z         |                                   R         T                                   *         ,  `         b             ̤         Τ             8         :  n         p             ڪ         ܪ             F         H  |         ~                                    T         V                      µ             ,                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               9!         <!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 _
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     -)-    )-#;-    #;-I-    I-}-l    }--    ---P    ---    -b-h	    b-W-p
    W-e-
    e-۰-    ۰-E׳-L                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               =!         >!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             0         1  a         b                                                                                  &             ?  (          h  Q            z                                                                      L!         M!  }"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        T#  $         $  %         %  &         &  2(         4(  j)         l)  *         *  +         +  -         -  J.         L.  /         /  0         0  1         1  *3         ,3  b4         d4  5         5  6         6  
8         8  B9         D9  z:         |:  ;         ;  <         <  ">         $>  Z?         \?  @         @  A         A  C         C  :D         <D  rE         tE  F         F  G         G  I         I  SJ         TJ  K         K  L         L  M          M  N          4O  /P          lP  gQ          Q  R          R  S         T  KU         LU  V         V  W         W  X         X  +Z         ,Z  c[         d[  \         \  ]         ]  _         _  C`         D`  {a         |a  b         b  c         c  #e         $e  [f         \f  g         g  h         h  j         j  ;k         <k  sl         tl  m         m  n         n  p         p  Sq         Tq  r         r  s         s  t         t  3v         4v  kw         lw  x         x  y         y  {         {  K|         L|  }         }  ~         ~             +         ,  c         d             ӄ         Ԅ             C         D  {         |                        #         $  [         \             ˏ         ̏                                b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               E!         V!  1         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    x
         y
             j                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     %
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-s-     s--    -;;-x    ;;-I-|    I-#}-\    #}-A-    A--h    -{b-    {b--	    --h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               0!         4!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      .         /  _	         `	  
         
           g                                   *         +  [          \  @          }  a                                                          *            S  C           |   l!          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            2         4  f         h                                   6         8  j         l  	         	  
         
             :         <  n         p                        
           >         @  r         t                                   B         D  v         x                                   F         H  z          |   !         !  "         "  $         $  J%         L%  ~&         &  '         '  (         (  *         *  N+         P+  ,         ,  -         -  .         .  0          0  R1         T1  2         2  3         3  4         4  "6         $6  V7         X7  8         8  9         9  :         :  &<         (<  Z=         \=  >         >  ?         ?  @         @  *B         ,B  ^C         `C  D         D  E         E  F         F  .H         0H  bI         dI  J         J  K         K  L          M  2N         4N  fO         hO  P         P  Q         Q  S         S  6T         8T  jU         lU  V         W  QX         SX  Y         Y  Z         Z  [         [  !]         #]  U^         W^  _         _  `         `  a         a  %c         'c  Yd         [d  e         e  f         f  g         g  )i         +i  ]j          'x  y          My  9z          sz  _{          {  |          |  }         T                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               4!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     -)-    )-5-    5-;D-    ;D-w-l    w-U2-    U2--x    -b-	    b-Q-
    Q-_-
    _--|    -ֳ-    ֳ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ;!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         )            R  <          {  e                                                          0	  
          Q
  /         r                                   5         6  f           <         =  m         n                                                       b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |           {                                              0           "  #         #  !%         #%  [&         ]&  '         '  (         (  	*         *  C+         E+  },         ,  -         -  .         .  +0         -0  e1         g1  2         2  3         3  5         5  M6         O6  7         7  8         8  9         9  5;         7;  o<         q<  =         =  >         >  @         @  WA         YA  B         B  C         C  E         E  ?F         h  8i         :i  rj         tj  k         k  l         l   n         "n  Zo         \o  p         p  q         q  s         
s  Bt         Dt  |u         ~u  v         v  w         w  *y         ,y  dz         fz  {         {  |         |  ~         ~  L         N                                   4         6  n         p                                   V         X             ʌ         ̌             =         ?  w         y                        #         %  ]         _             ї         ӗ             E         G                                   -         /  g         i             ۢ         ݢ             O         Q             ç         ŧ             7         9  q         s                                 !  Y         [             Ͳ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               :!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    x
         y
             j      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    x
         y
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     m
            0     -,-X     ,-m-     m--    -5;-x    5;-I-|    I-}-\    }-A-    A--h    -ob-    ob--	    --h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               -!         2!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      .         /  _	         `	  
         
           g                                   *         +  [          \  A          }  b                                                          *            S  B           |   k!          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            3         4  g         h                                   7         8  k         l  	         	  
         
             ;         <  o         p                                   ?         @  s         t                                   C         D  w         x                                   G         H  {          |   !         !  "         "  $         $  K%         L%  &         &  '         '  (         (  *         *  O+         P+  ,         ,  -         -  .         .  0          0  S1         T1  2         2  3         3  4         4  #6         $6  W7         X7  8         8  9         9  :         :  '<         (<  [=         \=  >         >  ?         ?  @         @  +B         ,B  _C         `C  D         D  E         E  F         F  /H         0H  cI         dI  J         J  K         K  L          M  3N         4N  gO         hO  P         P  Q         Q  S         S  7T         8T  kU         lU  V         W  RX         SX  Y         Y  Z         Z  [         [  "]         #]  V^         W^  _         _  `         `  a         a  &c         'c  Zd         [d  e         e  f         f  g         g  *i         +i  ^j          }  ~                      1            W  C          }  i         T                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               0!         @!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     -)-    )-5-    5-;D-    ;D-w-l    w-U2-    U2--x    -{b-	    {b-Q-
    Q-_-
    _--|    -ֳ-    ֳ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ;!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         )            R  ;          {  d                                                          0	  
          Q
  0         r                                   5         6  f           <         =  m         n                                                       b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         X  C            k                                             "  #         #  "%         #%  \&         ]&  '         '  (         (  
*         *  D+         E+  ~,         ,  -         -  .         .  ,0         -0  f1         g1  2         2  3         3  5         5  N6         O6  7         7  8         8  9         9  6;         7;  p<         q<  =         =  >         >  @         @  XA         YA  B         B  C         C  E         E  @F         h  9i         :i  sj         tj  k         k  l         l  !n         "n  [o         \o  p         p  q         q  	s         
s  Ct         Dt  }u         ~u  v         v  w         w  +y         ,y  ez         fz  {         {  |         |  ~         ~  M         N                                   5         6  o         p                                   W         X             ˌ         ̌             >         ?  x         y                        $         %  ^         _             җ         ӗ             F         G                                   .         /  h         i             ܢ         ݢ             P         Q             ħ         ŧ             8         9  r         s                                  !  Z         [             β                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               8!         @!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    x
         y
             j      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    x
         y
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     n
            0     -,-X     ,--     -;-    ;-;-x    ;-I-|    I-}-\    }--    --h    -{b-    {b--	    --h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               G!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             0         1  a         b                                  r  b                                                          ?            X  1          q  J            c            |                                              L!         M!  }"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            0         2  b         d                                   *         ,  \         ^  	         	  
         
             $         &  V         X                                               P         R                                              J         L  |         ~                                   D          F   v!         x!  "         "  #          :  o;          ;  <          <  =          =  >          ?  ?         F  G         G  &I         (I  XJ         ZJ  K         K  L         L  M         M   O         "O  RP         TP  Q         Q  R         R  S         S  U         U  LV         NV  ~W         W  X         X  Y         Y  [         [  F\         H\  x]         z]  ^         ^  _         _  a         a  @b         Bb  rc         tc  d         d  e         e  g         
g  :h         <h  li         ni  j         j  k         k  m         m  5n         6n  go         ho  p         p  q         q  r         r  /t         0t  au         bu  v         v  w         w  x         x  )z         *z  [{         \{  |         |  }         }  ~         ~  #         $  U         V                                              O         P                                              I         J  {                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               7!         J!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-;-x    ;-I-|    I-}-\    }--    --h    -b-    b-'n-	    'n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               :!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             0         1  a         b                                                                                  &            ?  )          h  R            {                                                                      L!         M!  }"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ;         <  w         x                        +         ,  g         h             	         	             W         X                        
           F         H                                   6         8  r         t                        &         (  b         d                                     R!         T!  "         "  #         #  %          "<   =          @=  >          ^>  <?          |?  Z@          @  xA         :H  uI         vI  J         J  K         K  )M         *M  eN         fN  O         O  P         P  R         R  US         VS  T         T  U         U  	W         
W  EX         FX  Y         Y  Z         Z  [         [  5]         6]  q^         r^  _         _  `         `  %b         &b  ac         bc  d         d  e         e  g         g  Qh         Rh  i         i  j         j  l         l  Am         Bm  |n         ~n  o         o  p         p  0r         2r  ls         ns  t         t  u         u   w         "w  \x         ^x  y         y  z         z  |         |  L}         N}  ~         ~                         <         >  x         z                        ,         .  h         j                                   X         Z             Џ         ҏ             H                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ;!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     E
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -A-    A-;-x    ;-I-|    I-}-\    }--    --h    -{b-    {b--	    --h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               I!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             0         1  a         b                                  r  a                                                          ?            X  2          q  K            d            }                                              L!         M!  }"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            1         2  c         d                                   +         ,  ]         ^  	         	  
         
             %         &  W         X                                               Q         R                                              K         L  }         ~                                   E          F   w!         x!  "         "  #          ,@  A          LA  /B          lB  OC          C  oD          D  E         F  G         G  'I         (I  YJ         ZJ  K         K  L         L  M         M  !O         "O  SP         TP  Q         Q  R         R  S         S  U         U  MV         NV  W         W  X         X  Y         Y  [         [  G\         H\  y]         z]  ^         ^  _         _  a         a  Ab         Bb  sc         tc  d         d  e         e  	g         
g  ;h         <h  mi         ni  j         j  k         k  m         m  4n         6n  fo         ho  p         p  q         q  r         r  .t         0t  `u         bu  v         v  w         w  x         x  (z         *z  Z{         \{  |         |  }         }  ~         ~  "         $  T         V                                              N         P                                              H         J  z                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               5!         J!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-;-x    ;-I-|    I-}-\    }--    --h    -b-    b-'n-	    'n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               <!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             0         1  a         b                                                                                  &             ?  (          h  Q            z                                                                      L!         M!  }"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            :         <  v         x                        *         ,  f         h             	         	             V         X                                   G         H                                   7         8  s         t                        '         (  c         d                                     S!         T!  "         "  #         #  %          A  B          B  C          C  D          E  E          0F  G         :H  tI         vI  J         J  K         K  (M         *M  dN         fN  O         O  P         P  R         R  TS         VS  T         T  U         U  W         
W  DX         FX  Y         Y  Z         Z  [         [  4]         6]  p^         r^  _         _  `         `  $b         &b  `c         bc  d         d  e         e  g         g  Ph         Rh  i         i  j         j  l         l  @m         Bm  }n         ~n  o         o  p         p  1r         2r  ms         ns  t         t  u         u  !w         "w  ]x         ^x  y         y  z         z  |         |  M}         N}  ~         ~                        =         >  y         z                        -         .  i         j                                   Y         Z             я         ҏ             I                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ;!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     F
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    <     -,-p     ,--     -5-    5-y6-    y6-E-    E-ax-t    ax-u9-     u9-A-X    A--    -b-p	    b-K-x
    K-Y-
    Y--    -ֳ-\    ֳ--<                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               E!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     2            K  $          d  =          }  s                                  :	  0
          y
  o                                 J         K  {         |                        I         J  z         {                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h         D%  G&         I  J         J  4L         5L  wM         xM  N         N  O         O  @Q         AQ  R         R  S         S  	U         
U  LV         MV  W         W  X         X  Z         Z  X[         Y[  \         \  ]         ]  !_         "_  d`         e`  a         a  b         b  -d         .d  pe         qe  f         f  g         g  9i         :i  |j         }j  k         k  m          m  n         Fn  o         o  p         p  r         r  Qs         Rs  t         t  u         u  w         w  ]x         ^x  y         y  z         z  &|         '|  i}         j}  ~         ~             2         3  u         v                        >         ?             Ĉ         ň             J         K             Ѝ         э             V         W             ܒ         ݒ               #          c  f                      +          ,  /         o                        3         8  v         {                        ?         D             ť         ʥ             K         P             Ѫ         ֪             W         \             ݯ                     %  c         h                        ,         1  o         t                                           b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               D!         P!  1         1  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    x
         y
             j                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    x
         y
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            0     -,-X     ,--     -#-    #-5-x    5-5D-|    5D-w-\    w-3-    3--h    -b-    b-!n-	    !n--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               :!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         )            R  <          {  e                                                          0	  
          Q
  /         r                                   5         6  f           <         =  m         n                                                       b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            =         >  {         |                        5         6  s         t             	         	  -         .  k         l                        %         &  c         d                                   [         \                                   S         T                                   K          L   !         !  "         "  $         $  C%         G  H         H  0J         2J  nK         pK  L         L  M         M  (O         *O  fP         hP  Q         Q  R         R   T         "T  ^U         `U  V         V  W         W  Y         Y  VZ         XZ  [         [  \         \  ^         ^  N_         P_  `         `  a         a  c         
c  Fd         Hd  e         e  f         f   h         h  >i         @i  |j         ~j  k         k  l          pm  pn         n  o         o  *q         ,q  hr         jr  s         s  t         t  "v         $v  `w         bw  x         x  y         y  {         {  X|         Z|  }         }  ~         ~             P         R             ̃         ΃  
           H         J             Ĉ         ƈ             @         B  ~                                 8         :  v         x                                   4  4          t  t                               >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               :!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     E
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    <     -,-p     ,--     -;-    ;-y6-    y6-E-    E-ax-t    ax-u9-     u9-A-X    A--    -b-p	    b-K-x
    K-Y-
    Y--    -ֳ-\    ֳ--<                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               G!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     2            K  %          d  >          }  t                                  :	  1
          y
  p                                 J         K  {         |                        I         J  z         {                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        D%  &         I  J         J  3L         5L  vM         xM  N         N  O         O  ?Q         AQ  R         R  S         S  U         
U  KV         MV  W         W  X         X  Z         Z  W[         Y[  \         \  ]         ]   _         "_  c`         e`  a         a  b         b  ,d         .d  oe         qe  f         f  g         g  8i         :i  {j         }j  k         k  m         m  Bn         Fn  o         o  p         p  r         r  Ps         Rs  t         t  u         u  w         w  \x         ^x  y         y  z         z  %|         '|  h}         j}  ~         ~             1         3  t         v                        =         ?             È         ň             I         K             ύ         э             U         W             ے         ݒ              _         c                        *         ,  k          o  r                                  8  ;          {  ~                      @         D             ƥ         ʥ  	           L         P             Ҫ         ֪             X         \             ޯ           !         %  d         h                        -         1  p         t                                           b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               D!         P!  1         1  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    x
         y
             j                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    x
         y
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            0     -,-X     ,--     -)-    )-5-x    5-5D-|    5D-w-\    w-u-    u--h    -{b-    {b-'n-	    'n--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               <!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         )            R  ;          {  d                                                          0	  
          Q
  0         r                                   5         6  f           <         =  m         n                                                       b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            <         >  z         |                        4         6  r         t             	         	  ,         .  j         l                        $         &  b         d                                   Z         \                                   R         T                                   J          L   !         !  "         "  $         $  B%         G  H         H  1J         2J  oK         pK  L         L  M         M  )O         *O  gP         hP  Q         Q  R         R  !T         "T  _U         `U  V         V  W         W  Y         Y  WZ         XZ  [         [  \         \  ^         ^  O_         P_  `         `  a         a  	c         
c  Gd         Hd  e         e  f         f  h         h  ?i         @i  }j         ~j  k         k  l         n  o         o  +q         ,q  ir         jr  s         s  t         t  #v         $v  aw         bw  x         x  y         y  {         {  Y|         Z|  }         }  ~         ~             Q         R             ̓         ΃             I         J             ň         ƈ             A         B                                   9         :  w         x                                   4  4          t  t                                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               8!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     F
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-;-x    ;-I-|    I-|-\    |-7-    7--h    -b-    b-n-	    n--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               @!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             0         1  a         b                                  r  b                                                          ?            X  1          q  J            c            |                                              L!         M!  }"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            >         @  ~                                  >         @  ~                      	          
  >         @  ~                                  >         @  ~                                  >         @  ~                                  >         @  ~                                  >         @  ~             !         !  "          #  >$         @$  ~%           2   3          @3  @4          4  5          5  6           7   8         K  L         L  /N         0N  oO         pO  P         P  Q         Q  /S         0S  oT         pT  U         U  V         V  /X         0X  oY         pY  Z         Z  [         [  /]         0]  o^         p^  _         _  `         `  /b         0b  oc         pc  d         d  e         e  /g         0g  oh         ph  i         i  j         j  /l         0l  om         pm  n         n  o         o  /q         0q  or           2         4  r         t                        2         4  r         t                        2         4  r         t                        2         4  r         t                        2         4  r         t                        2         4  r         t                        2         4  r         t                        2                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               C!         V!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     +
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-<-x    <-J-|    J-}-\    }--    --h    -cb-    cb-x-	    x--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               <!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      8         9  i	         j	  
         
             7         8  h         i                                              ;  7          z  v                                  7            P  )           i   B!          !  ["          "  t#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ?         @                                    ?         @                        	          
  ?         @                                    ?         @                                    ?         @                                    ?         @                                    ?         @               !         !  "          #  ?$         @$  %         %  &         &  '          (  ?)         @)  *         *  +         +  ,          -  ?.         @.  /         /  0         0  1          2  ?3         @3  4         4  5         5  6          7  ?8         @8  9         9  :         :  ;          <  ?=         @=  >         >  ?         ?  @          A  ?B         @B  C         C  D         D  E          F  ?G         @G  H         H  I         I  J          K  >L         @L  ~M         M  N         N  O          P  >Q         @Q  ~R         R  S         S  T          U  >V         @V  ~W         W  X         X  Y          Z  >[         @[  ~\         \  ]         ]  ^          _  >`         @`  ~a         a  b         b  c          d  >e         @e  ~f         f  g         g  h          i  >j         @j  ~k         k  l         l  m          y  z          z  {          |  |          <}  (~          h~  T            ^         `                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               /!         >!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 s
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-;-x    ;-I-|    I-|-\    |-7-    7--h    -b-    b-n-	    n--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               @!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             0         1  a         b                                  r  a                                                          ?            X  2          q  K            d            }                                              L!         M!  }"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ?         @                                    ?         @                        	          
  ?         @                                    ?         @                                    ?         @                                    ?         @                                    ?         @               !         !  "          #  ?$         @$  %          @8  @9          9  :          :  ;           <   =          @=  @>         K  L         L  .N         0N  nO         pO  P         P  Q         Q  .S         0S  nT         pT  U         U  V         V  .X         0X  nY         pY  Z         Z  [         [  .]         0]  n^         p^  _         _  `         `  .b         0b  nc         pc  d         d  e         e  .g         0g  nh         ph  i         i  j         j  .l         0l  nm         pm  n         n  o         o  .q         0q  nr           3         4  s         t                        3         4  s         t                        3         4  s         t                        3         4  s         t                        3         4  s         t                        3         4  s         t                        3         4  s         t                        3                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               G!         Z!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     *
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-<-x    <-J-|    J-}-\    }--    --h    -ib-    ib-x-	    x--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               :!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      =         >  n	         o	  
         
             <         =  m         n                                               @  =            |                                  <            U  /           n   H!          !  a"          "  z#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            >         @  ~                                  >         @  ~                      	          
  >         @  ~                                  >         @  ~                                  >         @  ~                                  >         @  ~                                  >         @  ~             !         !  "          #  >$         @$  ~%         %  &         &  '          (  >)         @)  ~*         *  +         +  ,          -  >.         @.  ~/         /  0         0  1          2  >3         @3  ~4         4  5         5  6          7  >8         @8  ~9         9  :         :  ;          <  >=         @=  ~>         >  ?         ?  @          A  >B         @B  ~C         C  D         D  E          F  >G         @G  ~H         H  I         I  J          K  ?L         @L  M         M  N         N  O          P  ?Q         @Q  R         R  S         S  T          U  ?V         @V  W         W  X         X  Y          Z  ?[         @[  \         \  ]         ]  ^          _  ?`         @`  a         a  b         b  c          d  ?e         @e  f         f  g         g  h          i  ?j         @j  k         k  l         l  m                                    ؂                      D  0            _         `                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               1!         B!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 s
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-m6-x    m6-E-|    E-Ux-\    Ux--    --h    -ub-    ub-n-	    n--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               <!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     2            K  $          d  =          }  s                                  :	  0
          y
  o                                 J         K  {         |                        I         J  z         {                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            <         >  z         |                        4         6  r         t             	         	  ,         .  j         l                        $         &  b         d                                   Z         \                                   R         T                                   J          L   !         !  "         "  $         $  B%          =  =          4>  $?          d?  T@          @  A          A  B         H  !J         "J  _K         `K  L         L  M         M  O         O  WP         XP  Q         Q  R         R  T         T  OU         PU  V         V  W         W  	Y         
Y  GZ         HZ  [         [  \         \  ^         ^  ?_         @_  }`         ~`  a         a  b         b  7d         8d  ue         ve  f         f  g         g  /i         0i  mj         nj  k         k  l         l  'n         o  q         q  Er         Fr  s         s  t         t  u          v  =w         >w  {x         |x  y         y  z         z  5|         6|  s}         t}  ~         ~             -         .  k         l                        %         &  c         d             ߉                      [         \             ׎         ؎             S         T             ϓ         Г                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               :!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     +
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-m-     m--    -5-x    5-/D-|    /D-}w-\    }w-s-    s--h    -Eb-    Eb-}x-	    }x--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               .!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         )            R  <          {  e                                                          0	  
          Q
  /         r                                   5         6  f           <         =  m         n                                                       b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       	          L  ;          ~  m                     p&  '         '  (         (  8*         9*  {+         |+  ,         ,  .         .  D/         E/  0         0  1         1  3         3  P4         Q4  5         5  6         6  8         8  \9         ]9  :         :  ;         ;  %=         &=  h>         i>  ?         ?  @         @  1B         2B  tC         uC  D         D  E         E  =G         >G  H         H  I         I  K         K  IL         JL  M         M  N         N  P         P  QQ         VQ  R         R  S         S  U         U  ]V         bV  W         W  X         X  &Z         +Z  i[         n[  \         \  ]         ]  2_         7_  u`         z`  a         a  b          c  >d         Cd  e         e  f         f  h         h  Ji         Oi  j         j  k         k  m         m  Vn         [n  o         o  p         p  r         $r  bs         gs  t         t  u         u  +w         0w  nx         sx  y         y  z         z  7|         <|  z}         }  ~         ~              C         H             Ƀ         ΃             O         T             Ո         ڈ             [         `                        $         )  g         l                        0         5  s         x                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               &!         0!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 s
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-m6-x    m6-E-|    E-Ux-\    Ux--    --h    -{b-    {b-n-	    n--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               <!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     2            K  %          d  >          }  t                                  :	  1
          y
  p                                 J         K  {         |                        I         J  z         {                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            =         >  {         |                        5         6  s         t             	         	  -         .  k         l                        %         &  c         d                                   [         \                                   S         T                                   K          L   !         !  "         "  $         $  C%          B  C          $D  E          TE  DF          F  tG          G  H         H   J         "J  ^K         `K  L         L  M         M  O         O  VP         XP  Q         Q  R         R  T         T  NU         PU  V         V  W         W  Y         
Y  FZ         HZ  [         [  \         \   ^         ^  >_         @_  |`         ~`  a         a  b         b  6d         8d  te         ve  f         f  g         g  .i         0i  lj         nj  k         k  l         l  &n         o  q         q  Dr         Fr  s         s  t         t  u          v  <w         >w  zx         |x  y         y  z         z  4|         6|  r}         t}  ~         ~             ,         .  j         l                        $         &  b         d             މ                      Z         \             ֎         ؎             R         T             Γ         Г                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               <!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     *
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -E -X     E -հ-     հ-ܰ-    ܰ-/-p    /-;-T    ;-c-4    c-s-    s--8    -Kb-    Kb-}x-	    }x--8
    -3׳-
                   =         b      J      7   U   7   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ,!         >!  /1                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 64 })  outputElements  5184    input0Elements  31104   input1Elements  24576   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     4                           #         b      J      7   M     operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       @                         )            R  ;          {  d                                                         (	  X
         Y
                                 p                                   3                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     4                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  1728    input0Elements  1728    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       x                                  F   5!          x!  g"         p&  '         '  (         (  7*         9*  z+         |+  ,         ,   .         .  C/         E/  0         0  1         1  3         3  O4         Q4  5         5  6         6  8         8  [9         ]9  :         :  ;         ;  $=         &=  g>         i>  ?         ?  @         @  0B         2B  sC         uC  D         D  E         E  <G         >G  H         H  I         I  K         K  HL         JL  M         M  N         N  P         P  RQ         VQ  R         R  S         S  U         U  ^V         bV  W         W  X         X  'Z         +Z  j[         n[  \         \  ]         ]  3_         7_  v`         z`  a         a  b          c  ?d         Cd  e         e  f         f  h         h  Ki         Oi  j         j  k         k  m         m  Wn         [n  o         o  p         p   r         $r  cs         gs  t         t  u         u  ,w         0w  ox         sx  y         y  z         z  8|         <|  {}         }  ~         ~             D         H             ʃ         ΃             P         T             ֈ         ڈ             \         `                        %         )  h         l                        1         5  t         x                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               (!         4!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 s
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-#;-x    #;-I-|    I-}-\    }-5-    5--h    -b-    b-Yx-	    Yx--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               <!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      .         /  _	         `	  
         
           g                                   *         +  [          \  @          }  a                                                          *            S  C           |   l!          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            :         <  v         x                        *         ,  f         h             	         	             V         X                        
           F         H                                   6         8  r         t                        &         (  b         d                                     R!         T!  "         "  #         #  %         %  B&         D&  ~'         '  (         (  )         )  2+         4+  n,         p,  -         -  .         .  "0         $0  ^1         `1  2         2  3         3  5         5  N6         P6  7         7  8         8  :         :  >;         @;  z<         |<  =         =  >         >  .@         0@  jA         lA  B         B  C         C  E          E  ZF         \F  G         G  H         H  J         J  WK         XK  L         L  M         M  O         O  GP         HP  Q         Q  R         R  S         S  7U         8U  sV         tV  W         W  X         X  'Z         (Z  c[         d[  \         \  ]         ]  _         _  S`         T`  a         a  b         b  d         d  Ce         De  f         f  g         g  h         h  3j         4j  ok         pk  l         l  m         m  #o          o  p          q  	r          Fr  Gs          s  t          t  u                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               M!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 _
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-s-     s--    -<-x    <-J-|    J-}-\    }--    --P    -ib-    ib-c-	    c--D
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ;!         <!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      8         9  i	         j	  
         
             7         8  h         i                                              ;  7          z  v                                  7            P  )           i   B!          !  ["          "  t#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            1         2  c         d                                   +         ,  ]         ^  	         	  
         
             %         &  W         X                                               Q         R                                              K         L  }         ~                                   E          F   w!         x!  "         "  #         #  %         %  ?&         @&  q'         r'  (         (  )         )  +         +  9,         :,  k-         l-  .         .  /         /  1         1  32         42  e3         f3  4         4  5         5  6         6  -8         .8  _9         `9  :         :  ;         ;  <         <  '>         (>  Y?         Z?  @         @  A         A  B         B  !D         "D  SE         TE  F         F  G         G  H         H  J          J  K          XK  TL          L  M          M  N          O  P         ir  s         s  t         t  u         u  /w         1w  ax         cx  y         y  z         z  {         {  )}         +}  [~         ]~                                   #         %  U         W                                              O         Q                                              I                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ?!         @!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 c
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     --    -#;-x    #;-I-|    I-}-\    }-k-    k--h    -b-    b-Yx-	    Yx--h
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               :!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      .         /  _	         `	  
         
           g                                   *         +  [          \  A          }  b                                                          *            S  B           |   k!          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ;         <  w         x                        +         ,  g         h             	         	             W         X                                   G         H                                   7         8  s         t                        '         (  c         d                                     S!         T!  "         "  #         #  %         %  C&         D&  '         '  (         (  )         )  3+         4+  o,         p,  -         -  .         .  #0         $0  _1         `1  2         2  3         3  5         5  O6         P6  7         7  8         8  :         :  ?;         @;  {<         |<  =         =  >         >  /@         0@  kA         lA  B         B  C         C  E          E  [F         \F  G         G  H         H  J         J  VK         XK  L         L  M         M  
O         O  FP         HP  Q         Q  R         R  S         S  6U         8U  rV         tV  W         W  X         X  &Z         (Z  b[         d[  \         \  ]         ]  _         _  R`         T`  a         a  b         b  d         d  Be         De  ~f         f  g         g  h         h  2j         4j  nk         pk  l         l  m         m  "o           v  w          >w  ?x          |x  }y          y  z          z  {                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               Q!         R!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 _
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-m-     m--    -<-x    <-J-|    J-}-\    }--    --P    -ib-    ib-c-	    c--D
    --׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               8!         :!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      =         >  n	         o	  
         
             <         =  m         n                                               @  =            |                                  <            U  /           n   H!          !  a"          "  z#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            0         2  b         d                                   *         ,  \         ^  	         	  
         
             $         &  V         X                                               P         R                                              J         L  |         ~                                   D          F   v!         x!  "         "  #         #  %         %  >&         @&  p'         r'  (         (  )         )  +         +  8,         :,  j-         l-  .         .  /         /   1         1  22         42  d3         f3  4         4  5         5  6         6  ,8         .8  ^9         `9  :         :  ;         ;  <         <  &>         (>  X?         Z?  @         @  A         A  B         B   D         "D  RE         TE  F         F  G         G  H         H  J          HP  DQ          Q  R          R  S          S  T          8U  4V         ir  s         s  t         t  u         u  0w         1w  bx         cx  y         y  z         z  {         {  *}         +}  \~         ]~                                   $         %  V         W                                              P         Q                                              J                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ?!         @!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 c
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-7-     7--    -g6-x    g6-E-|    E-Ox-\    Ox-"-    "--h    -b-    b-m-	    m--h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     R                       L!  1          1  t=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     4            N  &          h  @            |                                  9	  3
          v
  p                                 K         L  ~                                 H         I  {         |                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            4         6  j         l                                   B         D  x         z  	         	  
         
             P         R                                   (         *  ^         `                                    6         8  l         n                                   D         F  z         |               !         !  #         #  R$          I  UJ         VJ  K         K  L         L  M         M  -O         .O  cP         dP  Q         Q  R         R  T         T  ;U         <U  qV         rV  W         W  X         X  Z         Z  I[         J[  \         \  ]         ]  ^         ^  !`         "`  Wa         Xa  b         b  c         c  d         d  /f         0f  eg         fg  h         h  i         i  k         k  =l         >l  sm         tm  n         n  o         o  q         q  Kr         Lr  s         s  t         t  u         u  #w         $w  Yx         Zx  y         y  z         z  {         {  1}         2}  g~         h~             Ӏ         Ԁ  	         
  ?         @  u         v                                   M         N                                   %         &  [         \             Ǒ          0            T  6          x  Z            ~                               >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               O!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     +
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,-C-     C--    -:-x    :-I-|    I-|-\    |--    --h    -?b-    ?b--	    --h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     Y             !          N!  1          1  ~=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             2         3  e         f                                                                                  *            D  2          k  Y                                             z                                     E!         F!  x"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            6         8  n         p                                   N         P             	         	  
         
  .         0  f         h                                   F         H  ~                                 &         (  ^         `                                   >         @  v         x                          "          "  V#         X#  $         $  %         %  &          '  7(         8(  o)         p)  *         *  +         +  -         -  O.         P.  /         /  0         0  1         1  /3         03  g4         h4  5         5  6         6  8         8  G9         H9  :         :  ;         ;  <         <  '>         (>  _?         `?  @         @  A         A  C         C  ?D         @D  wE         xE  F         F  G         G  I          Fc  (d          jd  Le          e  pf          f  g          g  h         o   q         q  8r         :r  ps         rs  t         t  u         u  w         w  Px         Rx  y         y  z         z  {         {  0}         2}  h~         j~             ؀         ڀ             H         J                                   (         *  `         b             Ћ         ҋ           
  @         B  x         z                                  "  X                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               -!         .!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,-7-     7--    -g6-x    g6-E-|    E-Ox-\    Ox-'-    '--h    -{b-    {b-m-	    m--h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     S                       L!  1          1  u=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     2  
          K  #          d  <          }  x                                  4	  /
          q
  l                                 F         G  y         z                        C         D  v         w                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            5         6  k         l                                   C         D  y         z  	         	  
         
             Q         R                                   )         *  _         `                                   7         8  m         n                                   E         F  {         |               !         !  #         #  S$          I  TJ         VJ  K         K  L         L  M         M  ,O         .O  bP         dP  Q         Q  R         R  T         T  :U         <U  pV         rV  W         W  X         X  Z         Z  H[         J[  ~\         \  ]         ]  ^         ^   `         "`  Va         Xa  b         b  c         c  d         d  .f         0f  dg         fg  h         h  i         i  k         k  <l         >l  rm         tm  n         n  o         o  q         q  Jr         Lr  s         s  t         t  u         u  "w         $w  Xx         Zx  y         y  z         z  {         {  0}         2}  f~         h~             Ҁ         Ԁ           
  >         @  t         v                                   L         N                                   $         &  Z         \             Ƒ            ƣ                      ,            P  2          t  V                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               K!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     *
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,-7-     7--    -:-x    :-I-|    I-|-\    |--    --h    -Kb-    Kb--	    --h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     X                       J!  1          1  w=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             2         3  e         f                                                                                  &            ?  ,          f  S            z                                 u                                     @!         A!  s"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            7         8  o         p                                   O         P             	         	  
         
  /         0  g         h                                   G         H                                   '         (  _         `                                   ?         @  w         x                          "          "  W#         X#  $         $  %         %  &          '  6(         8(  n)         p)  *         *  +         +  -         -  N.         P.  /         /  0         0  1         1  .3         03  f4         h4  5         5  6         6  8         8  F9         H9  ~:         :  ;         ;  <         <  &>         (>  ^?         `?  @         @  A         A  C         C  >D         @D  vE         xE  F         F  G         G  I          h  i          j   k          Bk  $l          fl  Hm          m  ln         o  q         q  9r         :r  qs         rs  t         t  u         u  w         w  Qx         Rx  y         y  z         z  {         {  1}         2}  i~         j~             ـ         ڀ             I         J                                   )         *  a         b             ы         ҋ  	         
  A         B  y         z                        !         "  Y                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               1!         2!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,-I-     I--    -);-x    );-I-|    I-}-\    }--    --h    -]b-    ]b-Mx-	    Mx--h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     V             !          T!  1          1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      (         )  [	         \	  
         
           q                        	         
  <         =  o          p  L            o                                                          F  *          m  Q              x!          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            2         4  f         h                                   6         8  j         l  	         	  
         
             :         <  n         p                        
           >         @  r         t                                   B         D  v         x                                   F         H  z          |   !         !  "         "  $         $  J%         L%  ~&         &  '         '  (         (  *         *  N+         P+  ,         ,  -         -  .         .  0          0  R1         T1  2         2  3         3  4         4  "6         $6  V7         X7  8         8  9         9  :         :  &<         (<  Z=         \=  >         >  ?         ?  @         @  *B         ,B  ^C         `C  D         D  E         E  F         F  .H         ^H  I         I  J         J  K         K  -M         .M  aN         bN  O         O  P         P  Q         Q  1S         2S  eT         fT  U         U  V         V  X         X  5Y         6Y  iZ         jZ  [         [  \         \  ^         ^  9_         :_  m`         n`  a         a  b         b  	d         
d  =e         >e  qf         rf  g         g  h         h  j         j  Ak         Bk  ul                      I  E                      ό  ˍ                               >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               8!         :!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 ]
  }      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,-=-     =--    -;-x    ;-kJ-|    kJ-}-\    }--    --P    -ub-    ub-c-	    c--D
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     Y                       H!  1          1  u=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      
           =	         >	  p
         q
                        :         ;  m         n                                               N  >            {                                              9   !          S!  +"          m"  E#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            9         :  s         t                        !         "  [         \             	         	  	         
  C         D  }         ~                        +         ,  e         f                                   M         N                                   5         6  o         p                        !         !  W"         X"  #         #  $         $  &         &  ?'         @'  y(         z(  )         )  *         *  ',         (,  a-         b-  .         .  /         /  1         1  I2         J2  3         3  4         4  5         5  17         27  k8         l8  9         9  :         :  <         <  S=         T=  >         >  ?         ?  A         A  ;B         <B  uC         vC  D         D  E         E  #G         $G  \H         vl  m         m  n         n  "p         $p  \q         ^q  r         r  s         s  
u         u  Dv         Fv  ~w         w  x         x  y         y  ,{         .{  f|         h|  }         }  ~         ~             N         P                      ă             6         8  p         r                                    X         Z             ̎          Z  [                                  #  $          f  g                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               B!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 a
                     
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,-C-     C--    -);-x    );-I-|    I-}-\    }-U-    U--h    -cb-    cb-Mx-	    Mx--h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     U                       P!  1          1  }=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      (         )  [	         \	  
         
           q                        	         
  <         =  o          p  M            p                                                          F  )          m  P              w!          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            3         4  g         h                                   7         8  k         l  	         	  
         
             ;         <  o         p                                   ?         @  s         t                                   C         D  w         x                                   G         H  {          |   !         !  "         "  $         $  K%         L%  &         &  '         '  (         (  *         *  O+         P+  ,         ,  -         -  .         .  0          0  S1         T1  2         2  3         3  4         4  #6         $6  W7         X7  8         8  9         9  :         :  '<         (<  [=         \=  >         >  ?         ?  @         @  +B         ,B  _C         `C  D         D  E         E  F         F  /H         ^H  I         I  J         J  K         K  ,M         .M  `N         bN  O         O  P         P  Q         Q  0S         2S  dT         fT  U         U  V         V   X         X  4Y         6Y  hZ         jZ  [         [  \         \  ^         ^  8_         :_  l`         n`  a         a  b         b  d         
d  <e         >e  pf         rf  g         g  h         h  j         j  @k         Bk  tl          U  Q                      ۑ  ג                                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               :!         >!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 ]
  }      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,-7-     7--    -;-x    ;-kJ-|    kJ-}-\    }-% -    % --P    -{b-    {b-c-	    c--D
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     X                       F!  1          1  q=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 B	         C	  u
         v
                        ?         @  r         s                                               S  D                                  
            #            <   !          U!  -"          n"  F#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            8         :  r         t                                  "  Z         \             	         	           
  B         D  |         ~                        *         ,  d         f                                   L         N                                   4         6  n         p                        !         !  V"         X"  #         #  $         $  &         &  >'         @'  x(         z(  )         )  *         *  &,         (,  `-         b-  .         .  /         /  1         1  H2         J2  3         3  4         4  5         5  07         27  j8         l8  9         9  :         :  <         <  R=         T=  >         >  ?         ?   A         A  :B         <B  tC         vC  D         D  E         E  "G         $G  ]H         vl  m         m  n         n  #p         $p  ]q         ^q  r         r  s         s  u         u  Ev         Fv  w         w  x         x  y         y  -{         .{  g|         h|  }         }  ~         ~             O         P             Ã         ă             7         8  q         r                                    Y         Z             ͎                                  /  0          r  s                               >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               D!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 a
                     
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,-s-     s--    -;-x    ;-I-|    I-|-\    |--    -Y-h    Y-a-    a-߂-	    ߂--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               1!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             2         3  e         f                                  r  V            }                                              5            O  '          i  A            [            u         z                                     E!         F!  x"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            B         C                                   N         O                        
         
  Z         [                        #         $  f         g                        /         0  r         s                        ;         <  ~                                 G         H                          "         "  S#         T#  $         $  %         %  '         '  [(         `(  )         )  *         *  $,         ),  g-         l-  .         .  /         /  01         51  s2         x2  3         3  4         4  <6         A6  7         7  8         8  :         
:  H;         M;  <         <  =         =  ?         ?  T@         Y@  A         A  B         B  D         "D  `E         eE  F         F  G         G  )I          I  J          J  K          K  L          M  N          GN  +O         l  m         m  o         o  Rp         Sp  q         q  r         r  t         t  ^u         _u  v         v  w         w  'y         (y  jz         kz  {         {  |         |  3~         4~  v         w                        ?         @             Ņ         ƅ           	  K         L             ъ         Ҋ             W         X             ݏ         ޏ            !  c           ؔ         ٔ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     H                       ,!  t1          1  :=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,-s-     s--    -;-    ;-I-    I-|-l    |--    --P    --    -Kb-h	    Kb-E-p
    E-S-
    S--    -3׳-L                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               5!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             2         3  e         f                                                                                  *            D  2          k  Y                                             z                                     E!         F!  x"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        T#  $         $  %         %  '         '  C(         D(  )         )  *         *  +         +  3-         4-  o.         p.  /         /  0         0  #2         $2  _3         `3  4         4  5         5  7         7  O8         P8  9         9  :         :  <         <  ?=         @=  {>         |>  ?         ?  @         @  /B         0B  kC         lC  D         D  E         E  G          G  [H         \H  I         J  K         K  8M         :M  tN         vN  O         O  P         P  (R         *R  dS         fS  T         T  U         U  W         W  TX         VX  Y         Y  Z         Z  \         
\  D]         F]  ^         ^  _         _  `         `  4b         6b  pc         rc  d         d  e         e  $g         &g  `h         bh  i         i  j         j  l         l  Pm         Rm  n         n  o         o  q         q  @r         Br  |s         ~s  t         t  u         u  0w         2w  lx         nx  y         y  z          z  {          "|  }          ^}  [~          ~              Ӏ           L         N             Ą         Ƅ              <         >  x         z                        ,         .  h         j                                   X         Z                                b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     Y            !          Z!  1          2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    z
         {
             p                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                      
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-m-     m--    -;-x    ;-I-|    I-|-\    |--    -_-h    _-a-    a-߂-	    ߂--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               /!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             2         3  e         f                                  r  U            |                                              5            N  &          g  ?            X            q         u                                     @!         A!  s"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            A         C                        
           M         O                        
         
  Y         [                        "         $  e         g                        .         0  q         s                        :         <  }                                 F         H                          "         "  R#         T#  $         $  %         %  '         '  \(         `(  )         )  *         *  %,         ),  h-         l-  .         .  /         /  11         51  t2         x2  3         3  4         4  =6         A6  7         7  8         8  :         
:  I;         M;  <         <  =         =  ?         ?  U@         Y@  A         A  B         B  D         "D  aE         eE  F         F  G         G  *I          oO  SP          P  {Q          Q  R          R  S          T  T         l  m         m  o         o  Qp         Sp  q         q  r         r  t         t  ]u         _u  v         v  w         w  &y         (y  iz         kz  {         {  |         |  2~         4~  u         w                        >         @             ą         ƅ           	  J         L             Њ         Ҋ             V         X             ܏         ޏ           !  b           ה         ٔ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     I                       0!  y1          1  A=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                   	      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,-m-     m--    -;-    ;-I-    I-|-l    |--    --P    --    -Kb-h	    Kb-E-p
    E-S-
    S--    -3׳-L                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               3!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             2         3  e         f                                                                                  &            ?  ,          f  S            z                                 u                                     @!         A!  s"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        T#  $         $  %         %  '         '  B(         D(  ~)         )  *         *  +         +  2-         4-  n.         p.  /         /  0         0  "2         $2  ^3         `3  4         4  5         5  7         7  N8         P8  9         9  :         :  <         <  >=         @=  z>         |>  ?         ?  @         @  .B         0B  jC         lC  D         D  E         E  G          G  ZH         \H  I         J  K         K  9M         :M  uN         vN  O         O  P         P  )R         *R  eS         fS  T         T  U         U  W         W  UX         VX  Y         Y  Z         Z  	\         
\  E]         F]  ^         ^  _         _  `         `  5b         6b  qc         rc  d         d  e         e  %g         &g  ah         bh  i         i  j         j  l         l  Qm         Rm  n         n  o         o  q         q  Ar         Br  }s         ~s  t         t  u         u  1w         2w  mx         nx  y         y  z         z  !|         "|  ]}         ^}  ~         ~                                   N  K                      Ƅ  Å                     >  y         z                        -         .  i         j                                   Y         Z                                b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     X            !          Z!  1          2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    z
         {
             p                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                      
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-m-     m--    -5;-x    5;-I-|    I-}-\    }-+-    +--P    -!b-    !b-c-	    c-1-D
    1-3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               .!         @!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      (         )  [	         \	  
         
           q                        	         
  <         =  o          p  L            o                                                          F  *          m  Q              x!          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            >         C                                   J         O                        
         
  V         [                                 $  b         g                        +         0  n         s                        7         <  z                                  C         H                          "         "  O#         T#  $         $  %         %  '         '  [(         `(  )         )  *         *  $,         ),  g-         l-  .         .  /         /  01         51  s2         x2  3         3  4         4  <6         A6  7         7  8         8  :         
:  H;         M;  <         <  =         =  ?         ?  T@         Y@  A         A  B         B  D          P  Q          Q  R          R  S          T  T          >U  V         ng  h         h  i         i  2k         7k  ul         zl  m         m  n          o  >p         Cp  q         q  r         r  t         t  Ju         Ou  v         v  w         w  y         y  Sz         Xz  {         {  |         |  ~         ~  ]         b                        &         +  l         m                        2         3  t         u                        =         >           4  r         w                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     K                       D!  1          1  a=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 o
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,-y-     y--    -5-    5-MD-    MD-w-l    w-A4-    A4--x    -ob-	    ob-E-
    E-S-
    S--|    -ֳ-    ֳ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               7!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         '            N  <          u  c                                              	            ,	  
          O
  5         r                        
           =         >  p            R         S                                                                 b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         p  ]                                                          (#  c$         CF  ~G         G  H         H  I         I  8K         :K  vL         xL  M         M  N         N  0P         2P  nQ         pQ  R         R  S         S  (U         *U  fV         hV  W         W  X         X   Z         "Z  ^[         `[  \         \  ]         ]  _         _  V`         X`  a         a  b         b  d         d  Ne         Pe  f         f  g         g  i         
i  Fj         Hj  k         k  l         l   n         n  >o         @o  |p         ~p  q         q  r         r  6t         8t  tu         vu  v         v  w         w  .y         0y  lz         nz  {         {  |         |  &~         (~  d         f                                    \         ^             ؆         چ             T         V             Ћ         ҋ             L         N             Ȑ         ʐ             C         E                                   9         ;  w         y                        1         3  o         q                        )         +  g         i                        !         #  _         a             ۪         ݪ             W         Y             ӯ         կ             O         Q             ˴                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               8!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    z
         {
             p      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    z
         {
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            0     -,-X     ,-g-     g--    -5;-x    5;-I-|    I-}-\    }-1-    1--P    -!b-    !b-c-	    c-1-D
    1-3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               -!         >!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      (         )  [	         \	  
         
           q                        	         
  <         =  o          p  M            p                                                          F  )          m  P              w!          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            ?         C                                   K         O                        
         
  W         [                                  $  c         g                        ,         0  o         s                        8         <  {                                 D         H                          "         "  P#         T#  $         $  %         %  '         '  \(         `(  )         )  *         *  %,         ),  h-         l-  .         .  /         /  11         51  t2         x2  3         3  4         4  =6         A6  7         7  8         8  :         
:  I;         M;  <         <  =         =  ?         ?  U@         Y@  A         A  B         B  D          aV  @W          W  cX          X  Y          Y  Z          Z  [         ng  h         h  i         i  3k         7k  vl         zl  m         m  n          o  ?p         Cp  q         q  r         r  t         t  Ku         Ou  v         v  w         w  y         y  Tz         Xz  {         {  |         |  ~         ~  ^         b                        '         +  k         m                        1         3  s         u                        <         >           4  s         w                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     L                       D!  1          1  b=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 p
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,-y-     y--    -5-    5-MD-    MD-w-l    w-G4-    G4--x    -ib-	    ib-E-
    E-S-
    S--|    -ֳ-    ֳ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               5!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         '            N  ;          u  b                                              	            ,	  
          O
  6         r                        
           =         >  p            R         S                                                                 b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         L  9          x  e                          !          !  "         (#  d$         CF  G         G  H         H  I         I  9K         :K  wL         xL  M         M  N         N  1P         2P  oQ         pQ  R         R  S         S  )U         *U  gV         hV  W         W  X         X  !Z         "Z  _[         `[  \         \  ]         ]  _         _  W`         X`  a         a  b         b  d         d  Oe         Pe  f         f  g         g  	i         
i  Gj         Hj  k         k  l         l  n         n  ?o         @o  }p         ~p  q         q  r         r  7t         8t  uu         vu  v         v  w         w  /y         0y  mz         nz  {         {  |         |  '~         (~  e         f                                    ]         ^             ن         چ             U         V             ы         ҋ             M         N             ɐ         ʐ             D         E                                   :         ;  x         y                        2         3  p         q                        *         +  h         i                        "         #  `         a             ܪ         ݪ             X         Y             ԯ         կ             P         Q             ̴                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               4!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    z
         {
             p      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    z
         {
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            4     -,-`     ,-m-     m--    -;-    ;-I-    I-|-d    |--    --p    -b-    b-K-
    K-Y-
    Y-簳-t    簳-9׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               /!         >!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             2         3  e         f                                  r  V            }                                              5            O  '          i  A            [            u         z                                     E!         F!  x"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                             "            P  ?         d#  $         $  %         %  "'         $'  b(         d(  )         )  *         *  ",         $,  b-         d-  .         .  /         /  "1         $1  b2         d2  3         3  4         4  "6         $6  b7         d7  8         8  9         9  ";         $;  b<         d<  =         =  >         >  "@         $@  bA         dA  B         B  C         C  "E         $E  bF         dF  G         G  H         H  "J         $J  cK         dK  L         L  M         M  #O         $O  cP         dP  Q         Q  R         R  #T         $T  cU         dU  V         V  W         W  #Y         $Y  cZ         dZ  [         [  \         \  #^         $^  c_         d_  `         `  a         a  #c         $c  cd         dd  e         e  f         f  #h         $h  ci         di  j         j  k         k  #m         $m  cn         dn  o         ds  t         t  u         u  "w         $w  bx         dx  y         y  z         z  "|         $|  b}         d}  ~         ~             "         $  b         d                        "         $  b         d                        "         $  b         d                        "         $  b         d                        "         $  b         d                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               7!         R!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    z
         {
             p      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     >
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-;-x    ;-I-|    I-|-\    |-2-    2--h    -b-    b-n-	    n--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               <!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             2         3  e         f                                                                                  *            D  2          k  Y                                             z                                     E!         F!  x"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ?         @                                    ?         @                        	          
  ?         @                                    ?         @                                    ?         @                                    ?         @                                    ?         @               !          !  "          (#  $          T$  F%          %  r&          &  '         F  H         H  GI         I  J         J  L         L  QM         SM  N         N  O         O  Q         Q  QR         SR  S         S  T         T  V         V  QW         SW  X         X  Y         Y  [         [  Q\         S\  ]         ]  ^         ^  `         `  Qa         Sa  b         b  c         c  e         e  Qf         Sf  g         g  h         h  j         j  Qk         Sk  l         l  m         m  o         o  p         p  #r         $r  cs           א         ؐ             W         X             ו         ؕ             W         X             ך         ؚ             W         X             ן         ؟             W         X             פ         ؤ             W         X             ש         ة             W         X             ׮         خ             W         X                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ?!         R!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     A
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,-y-     y--    -;-    ;-I-    I-|-d    |--    --p    -b-    b-K-
    K-Y-
    Y-ᰳ-t    ᰳ-9׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               3!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             2         3  e         f                                  r  U            |                                              5            N  &          g  ?            X            q         u                                     @!         A!  s"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         ~  m                                   !  !          6"  %#         d#  $         $  %         %  #'         $'  c(         d(  )         )  *         *  #,         $,  c-         d-  .         .  /         /  #1         $1  c2         d2  3         3  4         4  #6         $6  c7         d7  8         8  9         9  #;         $;  c<         d<  =         =  >         >  #@         $@  cA         dA  B         B  C         C  #E         $E  cF         dF  G         G  H         H  #J         $J  bK         dK  L         L  M         M  "O         $O  bP         dP  Q         Q  R         R  "T         $T  bU         dU  V         V  W         W  "Y         $Y  bZ         dZ  [         [  \         \  "^         $^  b_         d_  `         `  a         a  "c         $c  bd         dd  e         e  f         f  "h         $h  bi         di  j         j  k         k  "m         $m  bn         dn  o         ds  t         t  u         u  #w         $w  cx         dx  y         y  z         z  #|         $|  c}         d}  ~         ~             #         $  c         d                        #         $  c         d                        #         $  c         d                        #         $  c         d                        #         $  c         d                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               7!         R!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    z
         {
             p      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     =
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-;-x    ;-I-|    I-|-\    |-2-    2--h    -b-    b-n-	    n--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               <!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             2         3  e         f                                                                                  &            ?  ,          f  S            z                                 u                                     @!         A!  s"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            >         @  ~                                  >         @  ~                      	          
  >         @  ~                                  >         @  ~                                  >         @  ~                                  >         @  ~                                  >         @  ~             !          '  (          )  )          0*  "+          \+  N,          ,  z-         F  H         H  FI         I  J         J  L         L  RM         SM  N         N  O         O  Q         Q  RR         SR  S         S  T         T  V         V  RW         SW  X         X  Y         Y  [         [  R\         S\  ]         ]  ^         ^  `         `  Ra         Sa  b         b  c         c  e         e  Rf         Sf  g         g  h         h  j         j  Rk         Sk  l         l  m         m  o         o  p         p  "r         $r  bs           ֐         ؐ             V         X             ֕         ؕ             V         X             ֚         ؚ             V         X             ֟         ؟             V         X             ֤         ؤ             V         X             ֩         ة             V         X             ֮         خ             V         X                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               =!         R!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     B
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,-y-     y--    -6-    6-E-    E-gx-l    gx-8-    8--x    -ob-	    ob-Q-
    Q-_-
    _--|    -ֳ-    ֳ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               1!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     4            N  &          h  @            |                                  9	  3
          v
  p                                 K         L  ~                                 H         I  {         |                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       C  D                                             zI  J         J  K         K  2M         4M  pN         rN  O         O  P         P  *R         ,R  hS         jS  T         T  U         U  "W         $W  `X         bX  Y         Y  Z         Z  \         \  X]         Z]  ^         ^  _         _  a         a  Pb         Rb  c         c  d         d  
f         f  Hg         Jg  h         h  i         i  k         k  Al         Bl  m         m  n         n  o         o  9q         :q  wr         xr  s         s  t         t  1v         2v  ow         pw  x         x  y         y  ){         *{  g|         h|  }         }  ~         ~  !         "  _         `             ۃ         ܃             W         X             ӈ         Ԉ             O         P             ˍ         ̍  	         
  G         H             Ò         Ē             ?         @  }         ~                        7         8  u         v                        /         0  m         n                        '         (  e         f                                    ]         ^             ٬         ڬ             U         V             ѱ         ұ             L         N             ȶ         ޶             Y                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               2!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    z
         {
             p      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    z
         {
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     z
            0     -,-X     ,--     -/-    /-5-x    5-SD-|    SD-w-\    w-g&-    g&--h    -b-    b-!n-	    !n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               >!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         '            N  <          u  c                                              	            ,	  
          O
  5         r                        
           =         >  p            R         S                                                                 b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            B         C                                   N         O                        
         
  Z         [                        #         $  f         g                        /         0  r         s                        ;         <  ~                                 G         H                          "         "  S#         T#  $         $  %         %  '         '  [(         `(  )         )  *         *  $,         ),  g-         l-  .         .  /         /  01         51  s2         x2  3         3  4         4  <6         A6  7         7  8         8  :         
:  H;         M;  <         <  =         =  ?         ?  T@         Y@  A         A  B         B  D         "D  `E         eE  F         i  j         j  7l         <l  zm         m  n         n   p         p  Cq         Hq  r         r  s         s  u         u  Ov         Tv  w         w  x         x  z         z  [{         `{  |         |  }         }  $         )  g         l                        0         5  s         x                        <         A                      ǋ           
  H         M             ΐ         Ӑ             T         Y                        0  2          l  n                                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               B!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     A
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     --    -y6-    y6-E-    E-ax-l    ax-8-    8--x    -ob-	    ob-Q-
    Q-_-
    _--|    -ֳ-    ֳ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               5!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     2  
          K  #          d  <          }  x                                  4	  /
          q
  l                                 F         G  y         z                        C         D  v         w                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         O  P                        	          
            [  \         zI  J         J  K         K  3M         4M  qN         rN  O         O  P         P  +R         ,R  iS         jS  T         T  U         U  #W         $W  aX         bX  Y         Y  Z         Z  \         \  Y]         Z]  ^         ^  _         _  a         a  Qb         Rb  c         c  d         d  f         f  Ig         Jg  h         h  i         i  k         k  @l         Bl  ~m         m  n         n  o         o  8q         :q  vr         xr  s         s  t         t  0v         2v  nw         pw  x         x  y         y  ({         *{  f|         h|  }         }  ~         ~            "  ^         `             ڃ         ܃             V         X             ҈         Ԉ             N         P             ʍ         ̍           
  F         H                      Ē              >         @  |         ~                        6         8  t         v                        .         0  l         n                        &         (  d         f                                    \         ^             ج         ڬ             T         V             б         ұ             M         N             ɶ         ޶             X                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               4!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    z
         {
             p      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    z
         {
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     y
            0     -,-X     ,--     -/-    /-5-x    5-SD-|    SD-w-\    w-,-    ,--h    -b-    b-!n-	    !n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               >!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         '            N  ;          u  b                                              	            ,	  
          O
  6         r                        
           =         >  p            R         S                                                                 b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            A         C                        
           M         O                        
         
  Y         [                        "         $  e         g                        .         0  q         s                        :         <  }                                 F         H                          "         "  R#         T#  $         $  %         %  '         '  \(         `(  )         )  *         *  %,         ),  h-         l-  .         .  /         /  11         51  t2         x2  3         3  4         4  =6         A6  7         7  8         8  :         
:  I;         M;  <         <  =         =  ?         ?  U@         Y@  A         A  B         B  D         "D  aE         eE  F         i  j         j  8l         <l  {m         m  n         n  p         p  Dq         Hq  r         r  s         s  u         u  Pv         Tv  w         w  x         x  z         z  \{         `{  |         |  }         }  %         )  h         l                        1         5  t         x                        =         A             Ë         ǋ           
  I         M             ϐ         Ӑ             U         Y               "          \  ^                      Ԫ  ֫                               >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               >!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     B
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     --    -;-x    ;-I-|    I-|-\    |-'9-    '9--h    -{b-    {b-n-	    n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               2!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             2         3  e         f                                  r  V            }                                              5            O  '          i  A            [            u         z                                     E!         F!  x"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            :         <  v         x                        *         ,  f         h             	         	             V         X                        
           F         H                                   6         8  r         t                        &         (  b         d                                     R!         T!  "         "  #         #  %         %  C&         D&  '         '  (         (  )         )  3+         4+  o,         p,  -         -  .         .  #0         $0  _1         `1  2         2  3         3  5         5  O6         P6  7         7  8         8  :         :  ?;         @;  {<         |<  =         =  >         >  /@         0@  kA         lA  B         B  C         C  E          E  [F         \F  G          S  |T          T  U          U  V          "W  X          TX  DY         m  o         o  Op           ֕         ؕ             N         P             ƚ         Ț             >         @  z         |                        .         0  j         l                                    Z         \             Ҫ         Ԫ             J         L             ¯         į              :         <  v         x                        *         ,  f         h                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               5!         J!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     +
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -A-    A-;-x    ;-}J-|    }J-}-\    }-]6-    ]6--h    -cb-    cb-x-	    x--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               F!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      
           =	         >	  p
         q
                        :         ;  m         n                                               N  >            {                                              9   !          S!  +"          m"  E#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            B         C                                   N         O                        
         
  Z         [                        #         $  f         g                        /         0  r         s                        ;         <  ~                                 G         H                          "         "  S#         T#  $         $  %         F  G         G  -I         .I  pJ         qJ  K         K  L         L  9N         :N  |O         }O  P         P  R         R  ES         FS  T         T  U         U  W         W  QX         RX  Y         Y  Z         Z  \         \  ]]         ^]  ^         ^  _         _  &a         'a  ib         jb  c         c  d         d  2f         3f  ug         vg  h         h  i         i  >k         ?k  l         l  m         m  o         o  Fp         Kp  q         q  r         r  t         t  Ru         Wu  v         v  w         w  y          y  ^z         cz  {         {  |         |  '~         ,~  j         o                        3         8  v         {                        ?         D             Ŋ         ʊ             K                      @  &          p  V                      Щ           ұ             S         X             ٶ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               +!         :!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 q
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-;-x    ;-I-|    I-|-\    |-'9-    '9--h    -b-    b-n-	    n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               4!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             2         3  e         f                                  r  U            |                                              5            N  &          g  ?            X            q         u                                     @!         A!  s"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ;         <  w         x                        +         ,  g         h             	         	             W         X                                   G         H                                   7         8  s         t                        '         (  c         d                                     S!         T!  "         "  #         #  %         %  B&         D&  ~'         '  (         (  )         )  2+         4+  n,         p,  -         -  .         .  "0         $0  ^1         `1  2         2  3         3  5         5  N6         P6  7         7  8         8  :         :  >;         @;  z<         |<  =         =  >         >  .@         0@  jA         lA  B         B  C         C  E          E  ZF         \F  G          Y  vZ          Z  [          [  \          ]  ^          N^  >_         m  o         o  Np           ו         ؕ             O         P             ǚ         Ț             ?         @  {         |                        /         0  k         l                                    [         \             Ӫ         Ԫ             K         L             ï         į              ;         <  w         x                        +         ,  g         h                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               9!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     *
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -;-    ;-;-x    ;-J-|    J-}-\    }-]6-    ]6--h    -ib-    ib-x-	    x--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               B!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 B	         C	  u
         v
                        ?         @  r         s                                               S  D                                  
            #            <   !          U!  -"          n"  F#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            A         C                        
           M         O                        
         
  Y         [                        "         $  e         g                        .         0  q         s                        :         <  }                                 F         H                          "         "  R#         T#  $         $  %         F  G         G  ,I         .I  oJ         qJ  K         K  L         L  8N         :N  {O         }O  P         P  R         R  DS         FS  T         T  U         U  W         W  PX         RX  Y         Y  Z         Z  \         \  \]         ^]  ^         ^  _         _  %a         'a  hb         jb  c         c  d         d  1f         3f  tg         vg  h         h  i         i  =k         ?k  l         l  m         m  o         o  Gp         Kp  q         q  r         r  t         t  Su         Wu  v         v  w         w  y          y  _z         cz  {         {  |         |  (~         ,~  k         o                        4         8  w         {                        @         D             Ɗ         ʊ  	           L                       0            `  F            v                     ұ             T         X             ڶ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               -!         >!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 q
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-6-x    6-E-|    E-mx-\    mx--    --h    -b-    b-n-	    n--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               4!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     4            N  &          h  @                                              9	  8
          v
  u                                 K         L  ~                                 H         I  {         |                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            8         :  r         t                                  "  Z         \             	         	           
  B         D  |         ~                        *         ,  d         f                                   L         N                                   4         6  n         p                        !         !  V"         X"  #         #  $          ;  <          =  =          &>  ?          H?  (@          j@  JA         H  J         J  WK         XK  L         L  M         M  O         O  ?P         @P  yQ         zQ  R         R  S         S  'U         (U  aV         bV  W         W  X         X  Z         Z  I[         J[  \         \  ]         ]  ^         ^  1`         2`  ka         la  b         b  c         c  e         e  Sf         Tf  g         g  h         h  j         j  ;k         <k  ul         vl  m         No  p         p  q         q  r         r  5t         6t  ou         pu  v         v  w         w  y         y  Wz         Xz  {         {  |         |  ~         ~  ?         @  y         z                        '         (  a         b             Շ         և             I         J                                   1         2  k         l             ߒ                              >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               8!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     +
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-A;-x    A;-I-|    I-)}-\    )}--    --h    -Qb-    Qb-͂-	    ͂--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               :!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             2         3  e         f                                                                                  *            D  -          k  T            {                                                         5          6   h!         i!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            2         4  f         h                                   6         8  j         l  	         	  
         
             :         <  n         p                        
           >         @  r         t                                   B         D  v         x                                   F         H  z          |   !         !  "         "  $         $  %          &  3'         4'  g(         h(  )         )  *         *  ,         ,  7-         8-  k.         l.  /         /  0         0  2         2  ;3         <3  o4         p4  5         5  6         6  8         8  ?9         @9  s:         t:  ;         ;  <         <  >         >  C?         D?  w@         x@  A         A  B         B  D         D  GE         HE  {F         |F  G         G  H          K  L          L  M          N  O          FO  ,P          nP  TQ         "o  Tp         Vp  q         q  r         r  s         s  $u         &u  Xv         Zv  w         w  x         x  y         y  ({         *{  \|         ^|  }         }  ~         ~             ,         .  `         b             Ȅ         ʄ             0         2  d         f             ̊         Ί              4         6  h         j             А         Ґ             8                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               $!         ,!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-6-x    6-E-|    E-gx-\    gx--    --h    -{b-    {b-n-	    n--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               6!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     2  
          K  #          d  <          }  {                                  4	  2
          q
  o                                 F         G  y         z                        C         D  v         w                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            9         :  s         t                        !         "  [         \             	         	  	         
  C         D  }         ~                        +         ,  e         f                                   M         N                                   5         6  o         p                        !         !  W"         X"  #         #  $          A  lB          B  C          C  D          D  E          F  F         H  J         J  VK         XK  L         L  M         M  O         O  >P         @P  xQ         zQ  R         R  S         S  &U         (U  `V         bV  W         W  X         X  Z         Z  H[         J[  \         \  ]         ]  ^         ^  0`         2`  ja         la  b         b  c         c  e         e  Rf         Tf  g         g  h         h   j         j  :k         <k  tl         vl  m         No  p         p  q         q  r         r  4t         6t  nu         pu  v         v  w         w  y         y  Vz         Xz  {         {  |         |  ~         ~  >         @  x         z                        &         (  `         b             ԇ         և             H         J                                   0         2  j         l             ޒ                              >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               6!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     *
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-5;-x    5;-I-|    I-}-\    }--    --h    -Wb-    Wb-͂-	    ͂--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               6!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             2         3  e         f                                                                                  &            ?  '          f  N            u                                                         +          ,   ^!         _!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            3         4  g         h                                   7         8  k         l  	         	  
         
             ;         <  o         p                                   ?         @  s         t                                   C         D  w         x                                   G         H  {          |   !         !  "         "  $         $  %          &  2'         4'  f(         h(  )         )  *         *  ,         ,  6-         8-  j.         l.  /         /  0         0  2         2  :3         <3  n4         p4  5         5  6         6  
8         8  >9         @9  r:         t:  ;         ;  <         <  >         >  B?         D?  v@         x@  A         A  B         B  D         D  FE         HE  zF         |F  G         G  H          Q  |R          R  S          S  T          U  U          6V  W         "o  Up         Vp  q         q  r         r  s         s  %u         &u  Yv         Zv  w         w  x         x  y         y  ){         *{  ]|         ^|  }         }  ~         ~             -         .  a         b             Ʉ         ʄ             1         2  e         f             ͊         Ί             5         6  i         j             ѐ         Ґ             9                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               &!         0!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -M-    M-e;-x    e;-I-|    I-M}-\    M}--    --h    -b-    b-kx-	    kx--h
    -K׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               K!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      (         )  [	         \	  
         
           q                        	         
  <         =  o          p  H            r                                              B            i  B            i              !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            7         8  o         p                                   O         P             	         	  
         
  /         0  g         h                                   G         H                                   '         (  _         `                                   ?         @  w         x                          "          "  W#         X#  $         $  %         %  &          '  6(         8(  n)         p)  *         *  +         +  -         -  N.         P.  /         /  0         0  1         1  .3         03  f4         h4  5         5  6         6  8         8  F9         H9  ~:         :  ;         ;  <         <  &>         (>  ^?         `?  @         @  A         A  C         C  >D         @D  vE         xE  F         F  G         G  I         I  K         K  ML         NL  M         M  N         N  O         O  -Q         .Q  eR         fR  S         S  T         T  V         V  EW         FW  }X         ~X  Y         Y  Z         Z  %\         &\  ]]         ^]  ^         ^  _         _  a         a  =b         >b  uc         vc  d         d  e         e  g         g  Uh         Vh  i         i  j         j  k         k  5m         6m  mn          "o  p          ep  Vq          q  r          r  s          .t  u                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               <!         B!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 ]
  }      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-<-x    <-J-|    J-}-\    }-w-    w--P    -b-    b-c-	    c-=-D
    =-K׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               >!         @!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 L	         M	  
         
                        I         J  |         }                                                ]  L                                              .             H    !          b!  :"          |"  T#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            5         6  k         l                                   C         D  y         z  	         	  
         
             Q         R                                   )         *  _         `                                   7         8  m         n                                   E         F  {         |               !         !  #         #  S$         T$  %         %  &         &  '         '  +)         ,)  a*         b*  +         +  ,         ,  .         .  9/         :/  o0         p0  1         1  2         2  4         4  G5         H5  }6         ~6  7         7  8         8  :          :  U;         V;  <         <  =         =  >         >  -@         .@  cA         dA  B         B  C         C  E         E  ;F         <F  pG         rG  H         H  I          V  W          FW  <X          ~X  tY          Y  Z          Z  [         nn  o         o  p         p  r         r  Ds         Fs  zt         |t  u         u  v         v  x         x  Ry         Ty  z         z  {         {  |         |  *~         ,~  `         b             ́         ΁             8         :  n         p             ڇ         ܇             F         H  |         ~                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               6!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 a
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -A-    A-_;-x    _;-I-|    I-G}-\    G}--    --h    -b-    b-kx-	    kx--h
    -K׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               G!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      (         )  [	         \	  
         
           q                        	         
  <         =  o          p  H            q                                              =            d  <            c              !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            6         8  n         p                                   N         P             	         	  
         
  .         0  f         h                                   F         H  ~                                 &         (  ^         `                                   >         @  v         x                          "          "  V#         X#  $         $  %         %  &          '  7(         8(  o)         p)  *         *  +         +  -         -  O.         P.  /         /  0         0  1         1  /3         03  g4         h4  5         5  6         6  8         8  G9         H9  :         :  ;         ;  <         <  '>         (>  _?         `?  @         @  A         A  C         C  ?D         @D  wE         xE  F         F  G         G  I         I  K         K  LL         NL  M         M  N         N  O         O  ,Q         .Q  dR         fR  S         S  T         T  V         V  DW         FW  |X         ~X  Y         Y  Z         Z  $\         &\  \]         ^]  ^         ^  _         _  a         a  <b         >b  tc         vc  d         d  e         e  g         g  Th         Vh  i         i  j         j  k         k  4m         6m  ln          qu  bv          v  w          w  x          :y  +z          }z  n{                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               >!         F!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 ]
  }      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-;-x    ;-J-|    J-}-\    }-w-    w--P    -b-    b-c-	    c-=-D
    =-K׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               =!         >!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 G	         H	  z
         {
                        D         E  w         x                                     	          X  F                                              (              A   !          Z!  2"          s"  K#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            4         6  j         l                                   B         D  x         z  	         	  
         
             P         R                                   (         *  ^         `                                    6         8  l         n                                   D         F  z         |               !         !  #         #  R$         T$  %         %  &         &  '         '  *)         ,)  `*         b*  +         +  ,         ,  .         .  8/         :/  n0         p0  1         1  2         2  4         4  F5         H5  |6         ~6  7         7  8         8  :          :  T;         V;  <         <  =         =  >         >  ,@         .@  bA         dA  B         B  C         C  E         E  :F         <F  qG         rG  H         H  I          &\  ]          ^]  T^          ^  _          _  `          a  a         nn  o         o  p         p  r         r  Es         Fs  {t         |t  u         u  v         v  x         x  Sy         Ty  z         z  {         {  |         |  +~         ,~  a         b             ́         ΁             9         :  o         p             ۇ         ܇             G         H  }         ~                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               8!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 a
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-6-x    6-%E-|    %E-sx-\    sx-7-    7--h    -ob-    ob-kx-	    kx--h
    -K׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               C!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     4            N  &          h  @            t                                  9	  +
          v
  h                                 K         L  ~                                 H         I  {         |                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                 $            g  _                     U$  %         %  &         &  '         '  ,)         -)  b*         c*  +         +  ,         ,  .         .  :/         ;/  p0         q0  1         1  2         2  4         4  H5         I5  ~6         6  7         7  8         8   :         !:  V;         W;  <         <  =         =  >         >  .@         /@  dA         eA  B         B  C         C  E         E  <F         =F  rG         G  I         I  CJ         EJ  yK         {K  L         L  M         M  O         O  QP         SP  Q         Q  R         R  S         S  )U         +U  _V         aV  W         W  X         X  Z         Z  7[         9[  m\         o\  ]         ]  ^         ^  `         `  Ea         Ga  {b         }b  c         c  d         d  f         f  Sg         Ug  h         h  i         i  j         j  +l         l  
n         ړ             E         F  {         |                                   S         T                                   +         ,  a         b             ͤ         Τ             9         :  o         p             ۪         ܪ             G         H  }         ~                                    U         V                      µ             -                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               3!         4!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 ]
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -)-    )-#;-    #;-I-    I-}-d    }--    --p    -b-    b-]-
    ]-k-
    k-۰-t    ۰-K׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               =!         >!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             2         3  e         f                                                                                  *            D  *          k  Q            x                                 z                                     E!         F!  x"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         <  =                                              H  I         T#  $         $  %         %  &         &  3(         4(  k)         l)  *         *  +         +  -         -  K.         L.  /         /  0         0  1         1  +3         ,3  c4         d4  5         5  6         6  8         8  C9         D9  {:         |:  ;         ;  <         <  #>         $>  [?         \?  @         @  A         A  C         C  ;D         <D  sE         tE  F         F  G         G  I         I  RJ         TJ  K         K  L         L  M         M  2O         4O  jP         lP  Q         Q  R         R  T         T  JU         LU  V         V  W         W  X         X  *Z         ,Z  b[         d[  \         \  ]         ]  
_         _  B`         D`  za         |a  b         b  c         c  "e         $e  Zf         \f  g         g  h         h  j         j  :k         <k  rl         tl  m         m  n         n  p         p  Rq         Tq  r         r  s         s  t         t  2v         4v  jw         lw  x         x  y         y  {         {  J|         L|  }         }  ~         ~             *         ,  b         d             ҄         Ԅ  
           B         D  z         |                        "         $  Z         \             ʏ         ̏                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               A!         R!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    z
         {
             p      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     &
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-6-x    6-E-|    E-mx-\    mx-7-    7--h    -ub-    ub-kx-	    kx--h
    -K׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ?!         @!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     2  
          K  #          d  <          }  p                                  4	  '
          q
  d                                 F         G  y         z                        C         D  v         w                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                     0  (          s  k                                 U$  %         %  &         &  '         '  +)         -)  a*         c*  +         +  ,         ,  .         .  9/         ;/  o0         q0  1         1  2         2  4         4  G5         I5  }6         6  7         7  8         8  :         !:  U;         W;  <         <  =         =  >         >  -@         /@  cA         eA  B         B  C         C  E         E  ;F         =F  qG         G  I         I  DJ         EJ  zK         {K  L         L  M         M  O         O  RP         SP  Q         Q  R         R  S         S  *U         +U  `V         aV  W         W  X         X  Z         Z  8[         9[  n\         o\  ]         ]  ^         ^  `         `  Fa         Ga  |b         }b  c         c  d         d  f         f  Tg         Ug  h         h  i         i  j         j  ,l         l  	n         ړ             D         F  z         |                                   R         T                                   *         ,  `         b             ̤         Τ             8         :  n         p             ڪ         ܪ             F         H  |         ~                                    T         V                      µ             ,                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               5!         8!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 ]
  ~      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -5-    5-;-    ;-I-    I-}-d    }-#-    #--p    -b-    b-]-
    ]-k-
    k-۰-t    ۰-K׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               A!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             2         3  e         f                                                                                  &            ?  $          f  K            r                                 u                                     @!         A!  s"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                         !          "  #         T#  $         $  %         %  &         &  2(         4(  j)         l)  *         *  +         +  -         -  J.         L.  /         /  0         0  1         1  *3         ,3  b4         d4  5         5  6         6  
8         8  B9         D9  z:         |:  ;         ;  <         <  ">         $>  Z?         \?  @         @  A         A  C         C  :D         <D  rE         tE  F         F  G         G  I         I  SJ         TJ  K         K  L         L  M         M  3O         4O  kP         lP  Q         Q  R         R  T         T  KU         LU  V         V  W         W  X         X  +Z         ,Z  c[         d[  \         \  ]         ]  _         _  C`         D`  {a         |a  b         b  c         c  #e         $e  [f         \f  g         g  h         h  j         j  ;k         <k  sl         tl  m         m  n         n  p         p  Sq         Tq  r         r  s         s  t         t  3v         4v  kw         lw  x         x  y         y  {         {  K|         L|  }         }  ~         ~             +         ,  c         d             ӄ         Ԅ             C         D  {         |                        #         $  [         \             ˏ         ̏            T  U                               >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               A!         R!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    z
         {
             p      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     %
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-y-     y--    -M;-x    M;-I-|    I-5}-\    5}-A-    A--h    -ob-    ob--	    --h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               4!         8!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      (         )  [	         \	  
         
           q                        	         
  <         =  o          p  T            w                                                          F  2          m  Y              !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            2         4  f         h                                   6         8  j         l  	         	  
         
             :         <  n         p                        
           >         @  r         t                                   B         D  v         x                                   F         H  z          |   !         !  "         "  $         $  J%         L%  ~&         &  '         '  (         (  *         *  N+         P+  ,         ,  -         -  .         .  0          0  R1         T1  2         2  3         3  4         4  "6         $6  V7         X7  8         8  9         9  :         :  &<         (<  Z=         \=  >         >  ?         ?  @         @  *B         ,B  ^C         `C  D         D  E         E  F         F  .H         0H  bI         dI  J         J  K         K  L          M  2N         4N  fO         hO  P         P  Q         Q  S         S  6T         8T  jU         lU  V         W  QX         SX  Y         Y  Z         Z  [         [  !]         #]  U^         W^  _         _  `         `  a         a  %c         'c  Yd         [d  e         e  f         f  g         g  )i         +i  ]j                      Ʉ              ن                      ;  %         T                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               0!         @!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     -)-    )-5-    5-_D-    _D-w-l    w-U2-    U2--x    -b-	    b-W-
    W-e-
    e--|    -ֳ-    ֳ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ;!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         '            N  4          u  [                                              	            ,	  

          O
  -         r                        
           =         >  p            R         S                                                                 b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                        (            P  9          x  a                     "  #         #  !%         #%  [&         ]&  '         '  (         (  	*         *  C+         E+  },         ,  -         -  .         .  +0         -0  e1         g1  2         2  3         3  5         5  M6         O6  7         7  8         8  9         9  5;         7;  o<         q<  =         =  >         >  @         @  WA         YA  B         B  C         C  E         E  ?F         h  8i         :i  rj         tj  k         k  l         l   n         "n  Zo         \o  p         p  q         q  s         
s  Bt         Dt  |u         ~u  v         v  w         w  *y         ,y  dz         fz  {         {  |         |  ~         ~  L         N                                   4         6  n         p                                   V         X             ʌ         ̌             =         ?  w         y                        #         %  ]         _             ї         ӗ             E         G                                   -         /  g         i             ۢ         ݢ             O         Q             ç         ŧ             7         9  q         s                                 !  Y         [             Ͳ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               :!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    z
         {
             p      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    z
         {
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     i
            0     -,-X     ,-s-     s--    -M;-x    M;-I-|    I-5}-\    5}-A-    A--h    -ib-    ib--	    --h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               1!         6!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      (         )  [	         \	  
         
           q                        	         
  <         =  o          p  U            x                                                
          F  1          m  X              !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            3         4  g         h                                   7         8  k         l  	         	  
         
             ;         <  o         p                                   ?         @  s         t                                   C         D  w         x                                   G         H  {          |   !         !  "         "  $         $  K%         L%  &         &  '         '  (         (  *         *  O+         P+  ,         ,  -         -  .         .  0          0  S1         T1  2         2  3         3  4         4  #6         $6  W7         X7  8         8  9         9  :         :  '<         (<  [=         \=  >         >  ?         ?  @         @  +B         ,B  _C         `C  D         D  E         E  F         F  /H         0H  cI         dI  J         J  K         K  L          M  3N         4N  gO         hO  P         P  Q         Q  S         S  7T         8T  kU         lU  V         W  RX         SX  Y         Y  Z         Z  [         [  "]         #]  V^         W^  _         _  `         `  a         a  &c         'c  Zd         [d  e         e  f         f  g         g  *i         +i  ^j          a  K            q                      ӌ                       T                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ,!         <!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     -/-    /-5-    5-_D-    _D-w-l    w-U2-    U2--x    -b-	    b-W-
    W-e-
    e--|    -ֳ-    ֳ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ?!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         '            N  3          u  Z                                              	            ,	  
          O
  .         r                        
           =         >  p            R         S                                                                 b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                   	          @	  )
          h
  Q         "  #         #  "%         #%  \&         ]&  '         '  (         (  
*         *  D+         E+  ~,         ,  -         -  .         .  ,0         -0  f1         g1  2         2  3         3  5         5  N6         O6  7         7  8         8  9         9  6;         7;  p<         q<  =         =  >         >  @         @  XA         YA  B         B  C         C  E         E  @F         h  9i         :i  sj         tj  k         k  l         l  !n         "n  [o         \o  p         p  q         q  	s         
s  Ct         Dt  }u         ~u  v         v  w         w  +y         ,y  ez         fz  {         {  |         |  ~         ~  M         N                                   5         6  o         p                                   W         X             ˌ         ̌             >         ?  x         y                        $         %  ^         _             җ         ӗ             F         G                                   .         /  h         i             ܢ         ݢ             P         Q             ħ         ŧ             8         9  r         s                                  !  Z         [             β                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               <!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    z
         {
             p      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    z
         {
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     j
            4     -,-`     ,--     -/-    /-;-    ;-I-    I-|-d    |--    --p    -b-    b-Q-
    Q-_-
    _--t    -?׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               C!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             2         3  e         f                                  r  ^                                                          5            O  '          i  A            [            u         z                                     E!         F!  x"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                        2  %          d  W                                 %  F&         K&  '         '  (         (  *         *  R+         W+  ,         ,  -         -  /          /  ^0         c0  1         1  2         2  '4         ,4  j5         o5  6         6  7         7  39         89  v:         {:  ;         ;  <         =  ?>         D>  ?         ?  @         @  B         B  KC         PC  D         D  E         E  G         G  WH         \H  I         I  J         J   L         %L  cM         hM  N         N  O         O  ,Q         1Q  oR         tR  S         S  T         T  8V         =V  {W         W  X         X  Z         Z  D[         I[  \         \  ]         ]  _         _  P`         U`  a         a  b         b  d         d  \e         ae  f         f  g         g  %i         *i  hj         mj  k         k  l         l  5n         6n  xo         yo  p         p  q         q  As         Bs  t         t  u         u  
w         w  Mx         Nx  y         y  z         z  |         |  Y}         Z}  ~         ~             "         #  e         f                        .         /  q         r                        :         ;  }         ~                        F         G             ̒         ͒             N         S                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               7!         J!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    z
         {
             p      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     F
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-;-x    ;-I-|    I-|-\    |--    --h    -b-    b-'n-	    'n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               6!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             2         3  e         f                                                                                  *            D  *          k  Q            x                                 z                                     E!         F!  x"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ;         <  w         x                        +         ,  g         h             	         	             W         X                        
           F         H                                   6         8  r         t                        &         (  b         d                                     R!         T!  "         "  #         #  %          0  1          2  2          23  4          P4  ,5          n5  J6         :H  uI         vI  J         J  K         K  )M         *M  eN         fN  O         O  P         P  R         R  US         VS  T         T  U         U  	W         
W  EX         FX  Y         Y  Z         Z  [         [  5]         6]  q^         r^  _         _  `         `  %b         &b  ac         bc  d         d  e         e  g         g  Qh         Rh  i         i  j         j  l         l  Am         Bm  |n         ~n  o         o  p         p  0r         2r  ls         ns  t         t  u         u   w         "w  \x         ^x  y         y  z         z  |         |  L}         N}  ~         ~                         <         >  x         z                        ,         .  h         j                                   X         Z             Џ         ҏ             H                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               7!         J!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     A
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -5-    5-;-    ;-I-    I-|-d    |--    --p    -{b-    {b-Q-
    Q-_-
    _--t    -?׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               E!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             2         3  e         f                                  r  ]                                                          5            N  &          g  ?            X            q         u                                     @!         A!  s"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                     ,            ^  Q	          	  
          
           %  G&         K&  '         '  (         (  *         *  S+         W+  ,         ,  -         -  /          /  _0         c0  1         1  2         2  (4         ,4  k5         o5  6         6  7         7  49         89  w:         {:  ;         ;  <         =  @>         D>  ?         ?  @         @  	B         B  LC         PC  D         D  E         E  G         G  XH         \H  I         I  J         J  !L         %L  dM         hM  N         N  O         O  -Q         1Q  pR         tR  S         S  T         T  9V         =V  |W         W  X         X  Z         Z  E[         I[  \         \  ]         ]  _         _  Q`         U`  a         a  b         b  d         d  ]e         ae  f         f  g         g  &i         *i  ij         mj  k         k  l         l  4n         6n  wo         yo  p         p  q         q  @s         Bs  t         t  u         u  	w         w  Lx         Nx  y         y  z         z  |         |  X}         Z}  ~         ~             !         #  d         f                        -         /  p         r                        9         ;  |         ~                        E         G             ˒         ͒             O         S                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               5!         J!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    z
         {
             p      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     E
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-;-x    ;-I-|    I-|-\    |--    --h    -b-    b-'n-	    'n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               8!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             2         3  e         f                                                                                  &            ?  $          f  K            r                                 u                                     @!         A!  s"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            :         <  v         x                        *         ,  f         h             	         	             V         X                                   G         H                                   7         8  s         t                        '         (  c         d                                     S!         T!  "         "  #         #  %          6  h7          7  8          8  9          9  :          ;  ;         :H  tI         vI  J         J  K         K  (M         *M  dN         fN  O         O  P         P  R         R  TS         VS  T         T  U         U  W         
W  DX         FX  Y         Y  Z         Z  [         [  4]         6]  p^         r^  _         _  `         `  $b         &b  `c         bc  d         d  e         e  g         g  Ph         Rh  i         i  j         j  l         l  @m         Bm  }n         ~n  o         o  p         p  1r         2r  ms         ns  t         t  u         u  !w         "w  ]x         ^x  y         y  z         z  |         |  M}         N}  ~         ~                        =         >  y         z                        -         .  i         j                                   Y         Z             я         ҏ             I                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ;!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     B
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    <     -,-p     ,--     -/-    /-6-    6-E-    E-gx-t    gx-u9-     u9-A-X    A--    -b-p	    b-Q-x
    Q-_-
    _--    -ֳ-\    ֳ--<                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               A!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     4            N  &          h  @            t                                  9	  +
          v
  h                                 K         L  ~                                 H         I  {         |                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        D%  &         I  J         J  4L         5L  wM         xM  N         N  O         O  @Q         AQ  R         R  S         S  	U         
U  LV         MV  W         W  X         X  Z         Z  X[         Y[  \         \  ]         ]  !_         "_  d`         e`  a         a  b         b  -d         .d  pe         qe  f         f  g         g  9i         :i  |j         }j  k         k  m         m  An         Fn  o         o  p         p  r         r  Qs         Rs  t         t  u         u  w         w  ]x         ^x  y         y  z         z  &|         '|  i}         j}  ~         ~             2         3  u         v                        >         ?             Ĉ         ň             J         K             Ѝ         э             V         W             ܒ         ݒ              ^         c                        +         ,  j         o                        3         8  v         {                                    D  E                      ʥ  ˦           K         P             Ѫ         ֪             W         \             ݯ                     %  c         h                        ,         1  o         t                                           b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               @!         L!  1         1  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    z
         {
             p                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    z
         {
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            0     -,-X     ,--     --    -5-x    5-SD-|    SD-w-\    w-q-    q--h    -ub-    ub-!n-	    !n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               6!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         '            N  4          u  [                                              	            ,	  

          O
  -         r                        
           =         >  p            R         S                                                                 b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            =         >  {         |                        5         6  s         t             	         	  -         .  k         l                        %         &  c         d                                   [         \                                   S         T                                   K          L   !         !  "         "  $         $  C%          =  >          F>  2?          t?  `@          @  A          A  B         G  H         H  0J         2J  nK         pK  L         L  M         M  (O         *O  fP         hP  Q         Q  R         R   T         "T  ^U         `U  V         V  W         W  Y         Y  VZ         XZ  [         [  \         \  ^         ^  N_         P_  `         `  a         a  c         
c  Fd         Hd  e         e  f         f   h         h  >i         @i  |j         ~j  k         k  l         n  o         o  *q         ,q  hr         jr  s         s  t         t  "v         $v  `w         bw  x         x  y         y  {         {  X|         Z|  }         }  ~         ~             P         R             ̃         ΃  
           H         J             Ĉ         ƈ             @         B  ~                                 8         :  v         x                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               6!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     A
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    <     -,-p     ,--     -5-    5-y6-    y6-E-    E-ax-t    ax-u9-     u9-A-X    A--    -b-p	    b-Q-x
    Q-_-
    _--    -ֳ-\    ֳ--<                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               C!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     2  
          K  #          d  <          }  p                                  4	  '
          q
  d                                 F         G  y         z                        C         D  v         w                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        D%  &         I  J         J  3L         5L  vM         xM  N         N  O         O  ?Q         AQ  R         R  S         S  U         
U  KV         MV  W         W  X         X  Z         Z  W[         Y[  \         \  ]         ]   _         "_  c`         e`  a         a  b         b  ,d         .d  oe         qe  f         f  g         g  8i         :i  {j         }j  k         k  m         m  Bn         Fn  o         o  p         p  r         r  Ps         Rs  t         t  u         u  w         w  \x         ^x  y         y  z         z  %|         '|  h}         j}  ~         ~             1         3  t         v                        =         ?             È         ň             I         K             ύ         э             U         W             ے         ݒ              _         c                        *         ,  k         o                        4         8  w         {                        @         D             ƥ         ʥ  	                      P  Q                      ֪  ׫                     \             ޯ           !         %  d         h                        -         1  p         t                                           b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               @!         L!  1         1  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    z
         {
             p                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    z
         {
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            0     -,-X     ,--     -#-    #-5-x    5-SD-|    SD-w-\    w-q-    q--h    -ob-    ob-!n-	    !n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               8!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         '            N  3          u  Z                                              	            ,	  
          O
  .         r                        
           =         >  p            R         S                                                                 b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            <         >  z         |                        4         6  r         t             	         	  ,         .  j         l                        $         &  b         d                                   Z         \                                   R         T                                   J          L   !         !  "         "  $         $  B%          B  C          ,D  E          ZE  FF          F  tG         G  H         H  1J         2J  oK         pK  L         L  M         M  )O         *O  gP         hP  Q         Q  R         R  !T         "T  _U         `U  V         V  W         W  Y         Y  WZ         XZ  [         [  \         \  ^         ^  O_         P_  `         `  a         a  	c         
c  Gd         Hd  e         e  f         f  h         h  ?i         @i  }j         ~j  k         k  l          vm  bn         n  o         o  +q         ,q  ir         jr  s         s  t         t  #v         $v  aw         bw  x         x  y         y  {         {  Y|         Z|  }         }  ~         ~             Q         R             ̓         ΃             I         J             ň         ƈ             A         B                                   9         :  w         x                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               4!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     B
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-;-x    ;-I-|    I-|-\    |-7-    7--h    -b-    b-	n-	    	n--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               <!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             2         3  e         f                                  r  ^                                                          5            O  '          i  A            [            u         z                                     E!         F!  x"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            >         @  ~                                  >         @  ~                      	          
  >         @  ~                                  >         @  ~                                  >         @  ~                                  >         @  ~                                  >         @  ~             !         !  "          #  >$         @$  ~%          >  ~?          ?  @           A  A          @B  >C          C  ~D         K  L         L  /N         0N  oO         pO  P         P  Q         Q  /S         0S  oT         pT  U         U  V         V  /X         0X  oY         pY  Z         Z  [         [  /]         0]  o^         p^  _         _  `         `  /b         0b  oc         pc  d         d  e         e  /g         0g  oh         ph  i         i  j         j  /l         0l  om         pm  n         n  o         o  /q         0q  or           2         4  r         t                        2         4  r         t                        2         4  r         t                        2         4  r         t                        2         4  r         t                        2         4  r         t                        2         4  r         t                        2                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ?!         R!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     +
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-<-x    <-J-|    J-}-\    }--    --h    -ib-    ib-}x-	    }x--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               8!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                       2         3  e	         f	  
         
             /         0  b         c                                              9  1          v  n                                  -            G             a   9!          {!  S"          "  m#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ?         @                                    ?         @                        	          
  ?         @                                    ?         @                                    ?         @                                    ?         @                                    ?         @               !         !  "          #  ?$         @$  %         %  &         &  '          (  ?)         @)  *         *  +         +  ,          -  ?.         @.  /         /  0         0  1          2  ?3         @3  4         4  5         5  6          7  ?8         @8  9         9  :         :  ;          <  ?=         @=  >         >  ?         ?  @          A  ?B         @B  C         C  D         D  E          F  ?G         @G  H         H  I         I  J          K  >L         @L  ~M         M  N         N  O          P  >Q         @Q  ~R         R  S         S  T          U  >V         @V  ~W         W  X         X  Y          Z  >[         @[  ~\         \  ]         ]  ^          _  >`         @`  ~a         a  b         b  c          d  >e         @e  ~f         f  g         g  h          i  >j         @j  ~k         k  l         l  m           n  n          ,o  p          Xp  Bq          q  nr          r  s            ^         `                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               3!         B!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 q
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-;-x    ;-I-|    I-|-\    |-7-    7--h    -b-    b-n-	    n--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               <!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             2         3  e         f                                  r  ]                                                          5            N  &          g  ?            X            q         u                                     @!         A!  s"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ?         @                                    ?         @                        	          
  ?         @                                    ?         @                                    ?         @                                    ?         @                                    ?         @               !         !  "          #  ?$         @$  %          D  E           F  F          @G  >H          H  ~I          I  J         K  L         L  .N         0N  nO         pO  P         P  Q         Q  .S         0S  nT         pT  U         U  V         V  .X         0X  nY         pY  Z         Z  [         [  .]         0]  n^         p^  _         _  `         `  .b         0b  nc         pc  d         d  e         e  .g         0g  nh         ph  i         i  j         j  .l         0l  nm         pm  n         n  o         o  .q         0q  nr           3         4  s         t                        3         4  s         t                        3         4  s         t                        3         4  s         t                        3         4  s         t                        3         4  s         t                        3         4  s         t                        3                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               C!         V!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     *
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     --    -<-x    <-J-|    J-}-\    }--    --h    -ob-    ob-}x-	    }x--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               6!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      7         8  j	         k	  
         
             4         5  g         h                                               >  7          {  t                                  2  
          K  #           d   <!          }!  U"          "  n#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            >         @  ~                                  >         @  ~                      	          
  >         @  ~                                  >         @  ~                                  >         @  ~                                  >         @  ~                                  >         @  ~             !         !  "          #  >$         @$  ~%         %  &         &  '          (  >)         @)  ~*         *  +         +  ,          -  >.         @.  ~/         /  0         0  1          2  >3         @3  ~4         4  5         5  6          7  >8         @8  ~9         9  :         :  ;          <  >=         @=  ~>         >  ?         ?  @          A  >B         @B  ~C         C  D         D  E          F  >G         @G  ~H         H  I         I  J          K  ?L         @L  M         M  N         N  O          P  ?Q         @Q  R         R  S         S  T          U  ?V         @V  W         W  X         X  Y          Z  ?[         @[  \         \  ]         ]  ^          _  ?`         @`  a         a  b         b  c          d  ?e         @e  f         f  g         g  h          i  ?j         @j  k         k  l         l  m          s  t          u  u          4v  w          `w  Jx          x  vy            _         `                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               5!         F!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 q
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     --    -s6-x    s6-E-|    E-[x-\    [x--    --h    -ob-    ob-n-	    n--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               8!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     4            N  &          h  @            t                                  9	  +
          v
  h                                 K         L  ~                                 H         I  {         |                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            <         >  z         |                        4         6  r         t             	         	  ,         .  j         l                        $         &  b         d                                   Z         \                                   R         T                                   J          L   !         !  "         "  $         $  B%          %  &          '  '          *(  )          N)  **          r*  N+         H  !J         "J  _K         `K  L         L  M         M  O         O  WP         XP  Q         Q  R         R  T         T  OU         PU  V         V  W         W  	Y         
Y  GZ         HZ  [         [  \         \  ^         ^  ?_         @_  }`         ~`  a         a  b         b  7d         8d  ue         ve  f         f  g         g  /i         0i  mj         nj  k         k  l         l  'n         o  q         q  Er         Fr  s         s  t         t  u          v  =w         >w  {x         |x  y         y  z         z  5|         6|  s}         t}  ~         ~             -         .  k         l                        %         &  c         d             ߉                      [         \             ׎         ؎             S         T             ϓ         Г                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               6!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     +
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-s-     s--    -5-x    5-MD-|    MD-w-\    w-s-    s--h    -Qb-    Qb-wx-	    wx--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               0!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         '            N  4          u  [                                              	            ,	  

          O
  -         r                        
           =         >  p            R         S                                                                 b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                     &            X  E            w                     p&  '         '  (         (  8*         9*  {+         |+  ,         ,  .         .  D/         E/  0         0  1         1  3         3  P4         Q4  5         5  6         6  8         8  \9         ]9  :         :  ;         ;  %=         &=  h>         i>  ?         ?  @         @  1B         2B  tC         uC  D         D  E         E  =G         >G  H         H  I         I  K         K  IL         JL  M         M  N         N  P         P  QQ         VQ  R         R  S         S  U         U  ]V         bV  W         W  X         X  &Z         +Z  i[         n[  \         \  ]         ]  2_         7_  u`         z`  a         a  b          c  >d         Cd  e         e  f         f  h         h  Ji         Oi  j         j  k         k  m         m  Vn         [n  o         o  p         p  r         $r  bs         gs  t         t  u         u  +w         0w  nx         sx  y         y  z         z  7|         <|  z}         }  ~         ~              C         H             Ƀ         ΃             O         T             Ո         ڈ             [         `                        $         )  g         l                        0         5  s         x                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               *!         4!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 q
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     --    -s6-x    s6-E-|    E-[x-\    [x--    --h    -ob-    ob-n-	    n--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               8!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     2  
          K  #          d  <          }  p                                  4	  '
          q
  d                                 F         G  y         z                        C         D  v         w                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            =         >  {         |                        5         6  s         t             	         	  -         .  k         l                        %         &  c         d                                   [         \                                   S         T                                   K          L   !         !  "         "  $         $  C%          +  r,          ,  -          -  .          /  /          &0  1         H   J         "J  ^K         `K  L         L  M         M  O         O  VP         XP  Q         Q  R         R  T         T  NU         PU  V         V  W         W  Y         
Y  FZ         HZ  [         [  \         \   ^         ^  >_         @_  |`         ~`  a         a  b         b  6d         8d  te         ve  f         f  g         g  .i         0i  lj         nj  k         k  l         l  &n         o  q         q  Dr         Fr  s         s  t         t  u          v  <w         >w  zx         |x  y         y  z         z  4|         6|  r}         t}  ~         ~             ,         .  j         l                        $         &  b         d             މ                      Z         \             ֎         ؎             R         T             Γ         Г                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               8!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     *
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-m-     m--    -5-x    5-MD-|    MD-w-\    w-s-    s--h    -Wb-    Wb-wx-	    wx--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               .!         @!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         '            N  3          u  Z                                              	            ,	  
          O
  .         r                        
           =         >  p            R         S                                                                 b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                  R  ?            q                     p&  '         '  (         (  7*         9*  z+         |+  ,         ,   .         .  C/         E/  0         0  1         1  3         3  O4         Q4  5         5  6         6  8         8  [9         ]9  :         :  ;         ;  $=         &=  g>         i>  ?         ?  @         @  0B         2B  sC         uC  D         D  E         E  <G         >G  H         H  I         I  K         K  HL         JL  M         M  N         N  P         P  RQ         VQ  R         R  S         S  U         U  ^V         bV  W         W  X         X  'Z         +Z  j[         n[  \         \  ]         ]  3_         7_  v`         z`  a         a  b          c  ?d         Cd  e         e  f         f  h         h  Ki         Oi  j         j  k         k  m         m  Wn         [n  o         o  p         p   r         $r  cs         gs  t         t  u         u  ,w         0w  ox         sx  y         y  z         z  8|         <|  {}         }  ~         ~             D         H             ʃ         ΃             P         T             ֈ         ڈ             \         `                        %         )  h         l                        1         5  t         x                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ,!         8!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 q
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     --    -5;-x    5;-I-|    I-}-\    }--    --h    -b-    b-Sx-	    Sx--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               8!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      (         )  [	         \	  
         
           q                        	         
  <         =  o          p  T            w                                                          F  2          m  Y              !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            :         <  v         x                        *         ,  f         h             	         	             V         X                        
           F         H                                   6         8  r         t                        &         (  b         d                                     R!         T!  "         "  #         #  %         %  B&         D&  ~'         '  (         (  )         )  2+         4+  n,         p,  -         -  .         .  "0         $0  ^1         `1  2         2  3         3  5         5  N6         P6  7         7  8         8  :         :  >;         @;  z<         |<  =         =  >         >  .@         0@  jA         lA  B         B  C         C  E          E  ZF         \F  G         G  H         H  J         J  WK         XK  L         L  M         M  O         O  GP         HP  Q         Q  R         R  S         S  7U         8U  sV         tV  W         W  X         X  'Z         (Z  c[         d[  \         \  ]         ]  _         _  S`         T`  a         a  b         b  d         d  Ce         De  f         f  g         g  h         h  3j         4j  ok         pk  l         l  m         m  #o          6|  5}          t}  s~          ~                        .  -                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               I!         J!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 ]
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-g-     g--    -;-x    ;-J-|    J-}-\    }-7-    7--h    -ob-    ob--	    --h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               7!         8!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                       2         3  e	         f	  
         
             /         0  b         c                                              9  1          v  n                                  -            G             a   9!          {!  S"          "  m#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            B         C                                   N         O                        
         
  Z         [                        #         $  f         g                        /         0  r         s                        ;         <  ~                                 G         H                          "         "  S#         T#  $         $  %         %  '         '  _(         `(  )         )  *         *  (,         ),  k-         l-  .         .  /         /  41         51  w2         x2  3         3  4         4  @6         A6  7         7  8         8  	:         
:  L;         M;  <         <  =         =  ?         ?  X@         Y@  A         A  B         B  !D         "D  dE         eE  F         F  G         G  -I         .I  pJ         qJ  K          b  c          &d  e          Pe  8f          zf  bg          g  h         n  o         o  !q         &q  dr         	  G         L             ͙         ҙ             S         X             ٞ         ޞ           !  _         d                        (         -  k         p                        4         9  w         |                        @         E             Ʊ         ˱  	           L         Q             Ҷ         ׶                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ?!         @!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-y-     y--    -5;-x    5;-I-|    I-}-\    }--    --h    -b-    b-Sx-	    Sx--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               6!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      (         )  [	         \	  
         
           q                        	         
  <         =  o          p  U            x                                                
          F  1          m  X              !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ;         <  w         x                        +         ,  g         h             	         	             W         X                                   G         H                                   7         8  s         t                        '         (  c         d                                     S!         T!  "         "  #         #  %         %  C&         D&  '         '  (         (  )         )  3+         4+  o,         p,  -         -  .         .  #0         $0  _1         `1  2         2  3         3  5         5  O6         P6  7         7  8         8  :         :  ?;         @;  {<         |<  =         =  >         >  /@         0@  kA         lA  B         B  C         C  E          E  [F         \F  G         G  H         H  J         J  VK         XK  L         L  M         M  
O         O  FP         HP  Q         Q  R         R  S         S  6U         8U  rV         tV  W         W  X         X  &Z         (Z  b[         d[  \         \  ]         ]  _         _  R`         T`  a         a  b         b  d         d  Be         De  ~f         f  g         g  h         h  2j         4j  nk         pk  l         l  m         m  "o          l  k                                  &  %          d  c                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               M!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 ]
  ~      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-g-     g--    -<-x    <-J-|    J-}-\    }-7-    7--h    -cb-    cb--	    --h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               4!         6!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      7         8  j	         k	  
         
             4         5  g         h                                               >  7          {  t                                  2  
          K  #           d   <!          }!  U"          "  n#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            A         C                        
           M         O                        
         
  Y         [                        "         $  e         g                        .         0  q         s                        :         <  }                                 F         H                          "         "  R#         T#  $         $  %         %  '         '  ^(         `(  )         )  *         *  ',         ),  j-         l-  .         .  /         /  31         51  v2         x2  3         3  4         4  ?6         A6  7         7  8         8  :         
:  K;         M;  <         <  =         =  ?         ?  W@         Y@  A         A  B         B   D         "D  cE         eE  F         F  G         G  ,I         .I  oJ         qJ  K          h  i          i  j          "k  
l          Ll  4m          vm  ^n         n  o         o  "q         &q  er         	  H         L             Ι         ҙ             T         X             ڞ         ޞ           !  `         d                        )         -  l         p                        5         9  x         |                        A         E             Ǳ         ˱  
           M         Q             Ӷ         ׶                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ;!         <!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-=-     =--    -s6-x    s6-E-|    E-[x-\    [x-u--    u---h    -b-    b-m-	    m--h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     R                       P!  1          1  z=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     8            T  ,          p  H                                              =	  3
          x
  n                                 Q         R                                   L         M                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            4         6  j         l                                   B         D  x         z  	         	  
         
             P         R                                   (         *  ^         `                                    6         8  l         n                                   D         F  z         |               !         !  #         #  R$          I  UJ         VJ  K         K  L         L  M         M  -O         .O  cP         dP  Q         Q  R         R  T         T  ;U         <U  qV         rV  W         W  X         X  Z         Z  I[         J[  \         \  ]         ]  ^         ^  !`         "`  Wa         Xa  b         b  c         c  d         d  /f         0f  eg         fg  h         h  i         i  k         k  =l         >l  sm         tm  n         n  o         o  q         q  Kr         Lr  s         s  t         t  u         u  #w         $w  Yx         Zx  y         y  z         z  {         {  1}         2}  g~         h~             Ӏ         Ԁ  	         
  ?         @  u         v                                   M         N                                   %         &  [         \             Ǒ            x                                              (                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               S!         T!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     +
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,-7-     7--    -:-x    :-I-|    I-|-\    |--    --h    -9b-    9b--	    --h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     U                       J!  1          1  t=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             4         5  i         j                                                                                  2  
          N  8          s  ]                                             p                                     C!         D!  x"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            6         8  n         p                                   N         P             	         	  
         
  .         0  f         h                                   F         H  ~                                 &         (  ^         `                                   >         @  v         x                          "          "  V#         X#  $         $  %         %  &          '  7(         8(  o)         p)  *         *  +         +  -         -  O.         P.  /         /  0         0  1         1  /3         03  g4         h4  5         5  6         6  8         8  G9         H9  :         :  ;         ;  <         <  '>         (>  _?         `?  @         @  A         A  C         C  ?D         @D  wE         xE  F         F  G         G  I          W  X          Y  Y          &Z  [          J[  *\          n\  N]         o   q         q  8r         :r  ps         rs  t         t  u         u  w         w  Px         Rx  y         y  z         z  {         {  0}         2}  h~         j~             ؀         ڀ             H         J                                   (         *  `         b             Ћ         ҋ           
  @         B  x         z                                  "  X                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               )!         *!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,-=-     =--    -m6-x    m6-E-|    E-Ux-\    Ux-'3-    '3--h    -b-    b-m-	    m--h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     S                       P!  1          1  {=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     6            Q  )          l  D            ~                                  8	  /
          s
  j                                 L         M                                   G         H  |         }                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            5         6  k         l                                   C         D  y         z  	         	  
         
             Q         R                                   )         *  _         `                                   7         8  m         n                                   E         F  {         |               !         !  #         #  S$          I  TJ         VJ  K         K  L         L  M         M  ,O         .O  bP         dP  Q         Q  R         R  T         T  :U         <U  pV         rV  W         W  X         X  Z         Z  H[         J[  ~\         \  ]         ]  ^         ^   `         "`  Va         Xa  b         b  c         c  d         d  .f         0f  dg         fg  h         h  i         i  k         k  <l         >l  rm         tm  n         n  o         o  q         q  Jr         Lr  s         s  t         t  u         u  "w         $w  Xx         Zx  y         y  z         z  {         {  0}         2}  f~         h~             Ҁ         Ԁ           
  >         @  t         v                                   L         N                                   $         &  Z         \             Ƒ          L  ,          p  P            t                      ܲ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               O!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     *
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -{,-X     {,-1-     1--    -:-x    :-I-|    I-|-\    |--    --h    -?b-    ?b--	    --h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     T                       F!  1          1  m=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             4         5  i         j                                                                                  .            I  2          n  W            |                                 k                        	          
   >!         ?!  s"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            7         8  o         p                                   O         P             	         	  
         
  /         0  g         h                                   G         H                                   '         (  _         `                                   ?         @  w         x                          "          "  W#         X#  $         $  %         %  &          '  6(         8(  n)         p)  *         *  +         +  -         -  N.         P.  /         /  0         0  1         1  .3         03  f4         h4  5         5  6         6  8         8  F9         H9  ~:         :  ;         ;  <         <  &>         (>  ^?         `?  @         @  A         A  C         C  >D         @D  vE         xE  F         F  G         G  I          ]  r^          ^  _          _  `          `  a          "b  c         o  q         q  9r         :r  qs         rs  t         t  u         u  w         w  Qx         Rx  y         y  z         z  {         {  1}         2}  i~         j~             ـ         ڀ             I         J                                   )         *  a         b             ы         ҋ  	         
  A         B  y         z                        !         "  Y                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               -!         .!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,-O-     O--    -A;-x    A;-I-|    I-)}-\    )}-9-    9--h    -Wb-    Wb-ӂ-	    ӂ--h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     V            !          X!  1          2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      "         #  W	         X	  
         
           {                                   N         O              `                                                          =            b  B            g              !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            >         C                                   J         O                        
         
  V         [                                 $  b         g                        +         0  n         s                        7         <  z                                  C         H                          "         "  O#         T#  $         $  %         %  '         '  [(         `(  )         )  *         *  $,         ),  g-         l-  .         .  /         /  01         51  s2         x2  3         3  4         4  <6         A6  7         7  8         8  :         
:  H;         M;  <         <  =         =  ?         ?  T@         Y@  A         A  B         B  D         "D  `E         eE  F         F  G         G  )I         .I  lJ         qJ  K          o  p          q  q          :r  ,s          rs  dt          t  u         Z             ߖ           "         #  e         f                        .         /  q         r                        :         ;  }         ~                        F         G             ̩         ͩ             R         S             خ         ٮ             ^         _                        '         (  j         k                        3                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               4!         6!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -{,-X     {,-1-     1--    -;-x    ;-kJ-|    kJ-}-\    }-=-    =--P    -ob-    ob-c-	    c--D
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     U                       D!  1          1  k=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 9	         :	  n
         o
                         4         5  i         j                                               J  6            q                                              3   !          O!  '"          k"  C#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            9         :  s         t                        !         "  [         \             	         	  	         
  C         D  }         ~                        +         ,  e         f                                   M         N                                   5         6  o         p                        !         !  W"         X"  #         #  $         $  &         &  ?'         @'  y(         z(  )         )  *         *  ',         (,  a-         b-  .         .  /         /  1         1  I2         J2  3         3  4         4  5         5  17         27  k8         l8  9         9  :         :  <         <  S=         T=  >         >  ?         ?  A         A  ;B         <B  uC         vC  D         D  E         E  #G         $G  \H          n`  ^a          a  b          b  c          
d  d          >e  .f         vl  m         m  n         n  "p         $p  \q         ^q  r         r  s         s  
u         u  Dv         Fv  ~w         w  x         x  y         y  ,{         .{  f|         h|  }         }  ~         ~             N         P                      ă             6         8  p         r                                    X         Z             ̎                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               >!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 _
                     
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,-I-     I--    -;;-x    ;;-I-|    I-#}-\    #}-9-    9--h    -]b-    ]b-ӂ-	    ӂ--h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     U                       T!  1          1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      "         #  W	         X	  
         
           {                                   N         O              a                                                          =            b  A            f              !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ?         C                                   K         O                        
         
  W         [                                  $  c         g                        ,         0  o         s                        8         <  {                                 D         H                          "         "  P#         T#  $         $  %         %  '         '  \(         `(  )         )  *         *  %,         ),  h-         l-  .         .  /         /  11         51  t2         x2  3         3  4         4  =6         A6  7         7  8         8  :         
:  I;         M;  <         <  =         =  ?         ?  U@         Y@  A         A  B         B  D         "D  aE         eE  F         F  G         G  *I         .I  mJ         qJ  K          u  v          w  x          Rx  Dy          y  |z          z  {         Z             ޖ           !         #  d         f                        -         /  p         r                        9         ;  |         ~                        E         G             ˩         ͩ             Q         S             ׮         ٮ             ]         _                        &         (  i         k                        2                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               6!         :!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -{,-X     {,-1-     1--    -;-x    ;-kJ-|    kJ-}-\    }-=-    =--P    -ub-    ub-c-	    c--D
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     T                       B!  1          1  g=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      	         
  >	         ?	  s
         t
                        9         :  n         o                                               O  <            w                                               6   !          Q!  )"          l"  D#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            8         :  r         t                                  "  Z         \             	         	           
  B         D  |         ~                        *         ,  d         f                                   L         N                                   4         6  n         p                        !         !  V"         X"  #         #  $         $  &         &  >'         @'  x(         z(  )         )  *         *  &,         (,  `-         b-  .         .  /         /  1         1  H2         J2  3         3  4         4  5         5  07         27  j8         l8  9         9  :         :  <         <  R=         T=  >         >  ?         ?   A         A  :B         <B  tC         vC  D         D  E         E  "G         $G  ]H          rf  bg          g  h          h  i          j  j          Bk  2l         vl  m         m  n         n  #p         $p  ]q         ^q  r         r  s         s  u         u  Ev         Fv  w         w  x         x  y         y  -{         .{  g|         h|  }         }  ~         ~             O         P             Ã         ă             7         8  q         r                                    Y         Z             ͎                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               @!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 _
                     
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,-y-     y--    -;-x    ;-I-|    I-|-\    |--    -S-h    S-a-    a-ق-	    ق--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               5!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             4         5  i         j                                  r  R            w                                              +            G            c  ;            W            s         p                                     C!         D!  x"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            B         C                                   N         O                        
         
  Z         [                        #         $  f         g                        /         0  r         s                        ;         <  ~                                 G         H                          "         "  S#         T#  $         $  %         %  '         '  [(         `(  )         )  *         *  $,         ),  g-         l-  .         .  /         /  01         51  s2         x2  3         3  4         4  <6         A6  7         7  8         8  :         
:  H;         M;  <         <  =         =  ?         ?  T@         Y@  A         A  B         B  D         "D  `E         eE  F         F  G         G  )I          7U  V          _V  AW          W  iX          X  Y          Y  Z         l  m         m  o         o  Rp         Sp  q         q  r         r  t         t  ^u         _u  v         v  w         w  'y         (y  jz         kz  {         {  |         |  3~         4~  v         w                        ?         @             Ņ         ƅ           	  K         L             ъ         Ҋ             W         X             ݏ         ޏ            !  c           ؔ         ٔ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     E                       *!  o1          1  4=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,-m-     m--    -;-    ;-I-    I-|-l    |--    --P    --    -Eb-h	    Eb-K-p
    K-Y-
    Y--    -3׳-L                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               1!         >!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             4         5  i         j                                                                                  2  
          N  8          s  ]                                             p                                     C!         D!  x"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        T#  $         $  %         %  '         '  C(         D(  )         )  *         *  +         +  3-         4-  o.         p.  /         /  0         0  #2         $2  _3         `3  4         4  5         5  7         7  O8         P8  9         9  :         :  <         <  ?=         @=  {>         |>  ?         ?  @         @  /B         0B  kC         lC  D         D  E         E  G          G  [H          \H  WI          J  K         K  8M         :M  tN         vN  O         O  P         P  (R         *R  dS         fS  T         T  U         U  W         W  TX         VX  Y         Y  Z         Z  \         
\  D]         F]  ^         ^  _         _  `         `  4b         6b  pc         rc  d         d  e         e  $g         &g  `h         bh  i         i  j         j  l         l  Pm         Rm  n         n  o         o  q          q  r          Br  =s          ~s  yt         t  u         u  0w         2w  lx         nx  y         y  z         z   |         "|  \}         ^}  ~         ~                        L         N             Ą         Ƅ              <         >  x         z                        ,         .  h         j                                   X         Z                                b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     U             !          V!  1          2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    |
         }
             v                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                      
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-s-     s--    -;-x    ;-I-|    I-|-\    |--    -Y-h    Y-a-    a-ӂ-	    ӂ--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               3!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             4         5  i         j                                  r  Q            v                                              +            F            a  9          |  T            o         k                        	          
   >!         ?!  s"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            A         C                        
           M         O                        
         
  Y         [                        "         $  e         g                        .         0  q         s                        :         <  }                                 F         H                          "         "  R#         T#  $         $  %         %  '         '  \(         `(  )         )  *         *  %,         ),  h-         l-  .         .  /         /  11         51  t2         x2  3         3  4         4  =6         A6  7         7  8         8  :         
:  I;         M;  <         <  =         =  ?         ?  U@         Y@  A         A  B         B  D         "D  aE         eE  F         F  G         G  *I          Z  [          '\  	]          O]  1^          w^  Y_          _  `         l  m         m  o         o  Qp         Sp  q         q  r         r  t         t  ]u         _u  v         v  w         w  &y         (y  iz         kz  {         {  |         |  2~         4~  u         w                        >         @             ą         ƅ           	  J         L             Њ         Ҋ             V         X             ܏         ޏ           !  b           ה         ٔ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     F                       .!  t1          1  ;=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,-g-     g--    -;-    ;-I-    I-|-l    |--    --P    --    -Eb-h	    Eb-K-p
    K-Y-
    Y--    -3׳-L                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               /!         >!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             4         5  i         j                                                                                  .            I  2          n  W            |                                 k                        	          
   >!         ?!  s"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        T#  $         $  %         %  '         '  B(         D(  ~)         )  *         *  +         +  2-         4-  n.         p.  /         /  0         0  "2         $2  ^3         `3  4         4  5         5  7         7  N8         P8  9         9  :         :  <         <  >=         @=  z>         |>  ?         ?  @         @  .B         0B  jC         lC  D         D  E         E  G          G  ZH         \H  I         J  K         K  9M         :M  uN         vN  O         O  P         P  )R         *R  eS         fS  T         T  U         U  W         W  UX         VX  Y         Y  Z         Z  	\         
\  E]         F]  ^         ^  _         _  `         `  5b         6b  qc         rc  d         d  e         e  %g         &g  ah         bh  i         i  j         j  l         l  Qm         Rm  n         n  o         o  q         q  Ar         Br  }s         ~s  t          t  u          u  v          2w  -x          nx  iy          y  z         z  !|         "|  ]}         ^}  ~         ~                        M         N             ń         Ƅ             =         >  y         z                        -         .  i         j                                   Y         Z                                b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     T                       V!  1          2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    |
         }
             v                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                      
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-s-     s--    -G;-x    G;-I-|    I-/}-\    /}-1-    1--P    -'b-    'b-c-	    c-1-D
    1-3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               2!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      "         #  W	         X	  
         
           {                                   N         O              `                                                          =            b  B            g              !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            >         C                                   J         O                        
         
  V         [                                 $  b         g                        +         0  n         s                        7         <  z                                  C         H                          "         "  O#         T#  $         $  %         %  '         '  [(         `(  )         )  *         *  $,         ),  g-         l-  .         .  /         /  01         51  s2         x2  3         3  4         4  <6         A6  7         7  8         8  :         
:  H;         M;  <         <  =         =  ?         ?  T@         Y@  A         A  B         B  D          \  \          3]  ^          V^  3_          y_  V`          `  ya         ng  h         h  i         i  2k         7k  ul         zl  m         m  n          o  >p         Cp  q         q  r         r  t         t  Ju         Ou  v         v  w         w  y         y  Sz         Xz  {         {  |         |  ~         ~  ]         b                        &         +  l         m                        2         3  t         u                        =         >           4  r         w                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     K                       H!  1          1  g=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 m
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,-s-     s--    -5-    5-kD-    kD-w-l    w-G4-    G4--x    -ib-	    ib-Q-
    Q-_-
    _--|    -ֳ-    ֳ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               3!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         %            J  4          o  Y            ~                                              (	  
          M
  3         r                                   E         F  z         4  h         i                                   <                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                             <  '          h  S         (#  c$         CF  ~G         G  H         H  I         I  8K         :K  vL         xL  M         M  N         N  0P         2P  nQ         pQ  R         R  S         S  (U         *U  fV         hV  W         W  X         X   Z         "Z  ^[         `[  \         \  ]         ]  _         _  V`         X`  a         a  b         b  d         d  Ne         Pe  f         f  g         g  i         
i  Fj         Hj  k         k  l         l   n         n  >o         @o  |p         ~p  q         q  r         r  6t         8t  tu         vu  v         v  w         w  .y         0y  lz         nz  {         {  |         |  &~         (~  d         f                                    \         ^             ؆         چ             T         V             Ћ         ҋ             L         N             Ȑ         ʐ             C         E                                   9         ;  w         y                        1         3  o         q                        )         +  g         i                        !         #  _         a             ۪         ݪ             W         Y             ӯ         կ             O         Q             ˴                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               4!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    |
         }
             v      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    |
         }
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            0     -,-X     ,-s-     s--    -G;-x    G;-I-|    I-/}-\    /}-1-    1--P    --b-    -b-c-	    c-1-D
    1-3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               1!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      "         #  W	         X	  
         
           {                                   N         O              a                                                          =            b  A            f              !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            ?         C                                   K         O                        
         
  W         [                                  $  c         g                        ,         0  o         s                        8         <  {                                 D         H                          "         "  P#         T#  $         $  %         %  '         '  \(         `(  )         )  *         *  %,         ),  h-         l-  .         .  /         /  11         51  t2         x2  3         3  4         4  =6         A6  7         7  8         8  :         
:  I;         M;  <         <  =         =  ?         ?  U@         Y@  A         A  B         B  D          a  b          b  c          d  d          (e  f          Kf  (g         ng  h         h  i         i  3k         7k  vl         zl  m         m  n          o  ?p         Cp  q         q  r         r  t         t  Ku         Ou  v         v  w         w  y         y  Tz         Xz  {         {  |         |  ~         ~  ^         b                        '         +  k         m                        1         3  s         u                        <         >           4  s         w                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     L                       H!  1          1  h=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 n
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,-s-     s--    -5-    5-kD-    kD-w-l    w-G4-    G4--x    -cb-	    cb-Q-
    Q-_-
    _--|    -ֳ-    ֳ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               1!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         %            J  3          o  X            }                                              (	  
          M
  4         r                                   E         F  z         4  h         i                                   <                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                                         D  /         (#  d$         CF  G         G  H         H  I         I  9K         :K  wL         xL  M         M  N         N  1P         2P  oQ         pQ  R         R  S         S  )U         *U  gV         hV  W         W  X         X  !Z         "Z  _[         `[  \         \  ]         ]  _         _  W`         X`  a         a  b         b  d         d  Oe         Pe  f         f  g         g  	i         
i  Gj         Hj  k         k  l         l  n         n  ?o         @o  }p         ~p  q         q  r         r  7t         8t  uu         vu  v         v  w         w  /y         0y  mz         nz  {         {  |         |  '~         (~  e         f                                    ]         ^             ن         چ             U         V             ы         ҋ             M         N             ɐ         ʐ             D         E                                   :         ;  x         y                        2         3  p         q                        *         +  h         i                        "         #  `         a             ܪ         ݪ             X         Y             ԯ         կ             P         Q             ̴                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               0!         @!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    |
         }
             v      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    |
         }
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            8     -,-h     ,-s-     s--    -;-    ;-I-    I-|-l    |--    --P    --    -b-h	    b-Q-p
    Q-_-
    _-ᰳ-    ᰳ-9׳-L                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               /!         >!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             4         5  i         j                                  r  R            w                                              +            G            c  ;            W            s         p                                     C!         D!  x"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        d#  $         $  %         %  "'         $'  b(         d(  )         )  *         *  ",         $,  b-         d-  .         .  /         /  "1         $1  b2         d2  3         3  4         4  "6         $6  b7         d7  8         8  9         9  ";         $;  b<         d<  =         =  >         >  "@         $@  bA         dA  B         B  C         C  "E         $E  bF         dF  G         G  H          H  I         $J  cK         dK  L         L  M         M  #O         $O  cP         dP  Q         Q  R         R  #T         $T  cU         dU  V         V  W         W  #Y         $Y  cZ         dZ  [         [  \         \  #^         $^  c_         d_  `         `  a         a  #c         $c  cd         dd  e         e  f         f  #h         $h  ci         di  j         j  k         k  #m         $m  cn         dn  o          ds  ct          t  u          u  v          $w  #x         dx  y         y  z         z  "|         $|  b}         d}  ~         ~             "         $  b         d                        "         $  b         d                        "         $  b         d                        "         $  b         d                        "         $  b         d                                b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ;!         V!  1         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    |
         }
             v                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     :
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-;-x    ;-I-|    I-|-\    |-2-    2--h    -b-    b-n-	    n--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               8!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             4         5  i         j                                                                                  2  
          N  8          s  ]                                             p                                     C!         D!  x"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ?         @                                    ?         @                        	          
  ?         @                                    ?         @                                    ?         @                                    ?         @                                    ?         @               !          ,8  
9          K9  ):          j:  H;          ;  g<          <  =         F  H         H  GI         I  J         J  L         L  QM         SM  N         N  O         O  Q         Q  QR         SR  S         S  T         T  V         V  QW         SW  X         X  Y         Y  [         [  Q\         S\  ]         ]  ^         ^  `         `  Qa         Sa  b         b  c         c  e         e  Qf         Sf  g         g  h         h  j         j  Qk         Sk  l         l  m         m  o         o  p         p  #r         $r  cs           א         ؐ             W         X             ו         ؕ             W         X             ך         ؚ             W         X             ן         ؟             W         X             פ         ؤ             W         X             ש         ة             W         X             ׮         خ             W         X                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ;!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     =
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,-s-     s--    -;-    ;-I-    I-|-l    |--    --P    --    -b-h	    b-Q-p
    Q-_-
    _-ᰳ-    ᰳ-9׳-L                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               /!         >!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             4         5  i         j                                  r  Q            v                                              +            F            a  9          |  T            o         k                        	          
   >!         ?!  s"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        d#  $         $  %         %  #'         $'  c(         d(  )         )  *         *  #,         $,  c-         d-  .         .  /         /  #1         $1  c2         d2  3         3  4         4  #6         $6  c7         d7  8         8  9         9  #;         $;  c<         d<  =         =  >         >  #@         $@  cA         dA  B         B  C         C  #E         $E  cF         dF  G         G  H         H  #J         $J  bK         dK  L         L  M         M  "O         $O  bP         dP  Q         Q  R         R  "T         $T  bU         dU  V         V  W         W  "Y         $Y  bZ         dZ  [         [  \         \  "^         $^  b_         d_  `         `  a         a  "c         $c  bd         dd  e         e  f         f  "h         $h  bi         di  j         j  k         k  "m         $m  bn         dn  o         ds  t         t  u         u  #w         $w  cx          dx  cy          y  z          z  {          $|  #}          d}  c~         ~             #         $  c         d                        #         $  c         d                        #         $  c         d                        #         $  c         d                        #         $  c         d                                b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ;!         V!  1         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    |
         }
             v                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     9
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-;-x    ;-I-|    I-|-\    |-2-    2--h    -{b-    {b-n-	    n--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               8!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             4         5  i         j                                                                                  .            I  2          n  W            |                                 k                        	          
   >!         ?!  s"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            >         @  ~                                  >         @  ~                      	          
  >         @  ~                                  >         @  ~                                  >         @  ~                                  >         @  ~                                  >         @  ~             !          =  >          >  ?          @  @          $A  B          CB  !C         F  H         H  FI         I  J         J  L         L  RM         SM  N         N  O         O  Q         Q  RR         SR  S         S  T         T  V         V  RW         SW  X         X  Y         Y  [         [  R\         S\  ]         ]  ^         ^  `         `  Ra         Sa  b         b  c         c  e         e  Rf         Sf  g         g  h         h  j         j  Rk         Sk  l         l  m         m  o         o  p         p  "r         $r  bs           ֐         ؐ             V         X             ֕         ؕ             V         X             ֚         ؚ             V         X             ֟         ؟             V         X             ֤         ؤ             V         X             ֩         ة             V         X             ֮         خ             V         X                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               9!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     >
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,-y-     y--    -6-    6-%E-    %E-sx-l    sx-8-    8--x    -ob-	    ob-W-
    W-e-
    e--|    -ֳ-    ֳ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               1!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     8            T  ,          p  H                                              =	  3
          x
  n                                 Q         R                                   L         M                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                 $  #          g  f                     zI  J         J  K         K  2M         4M  pN         rN  O         O  P         P  *R         ,R  hS         jS  T         T  U         U  "W         $W  `X         bX  Y         Y  Z         Z  \         \  X]         Z]  ^         ^  _         _  a         a  Pb         Rb  c         c  d         d  
f         f  Hg         Jg  h         h  i         i  k         k  Al         Bl  m         m  n         n  o         o  9q         :q  wr         xr  s         s  t         t  1v         2v  ow         pw  x         x  y         y  ){         *{  g|         h|  }         }  ~         ~  !         "  _         `             ۃ         ܃             W         X             ӈ         Ԉ             O         P             ˍ         ̍  	         
  G         H             Ò         Ē             ?         @  }         ~                        7         8  u         v                        /         0  m         n                        '         (  e         f                                    ]         ^             ٬         ڬ             U         V             ѱ         ұ             L         N             ȶ         ޶             Y                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               2!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    |
         }
             v      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    |
         }
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     v
            0     -,-X     ,--     -)-    )-5-x    5-qD-|    qD-w-\    w--    --h    -b-    b-n-	    n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               :!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         %            J  4          o  Y            ~                                              (	  
          M
  3         r                                   E         F  z         4  h         i                                   <                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            B         C                                   N         O                        
         
  Z         [                        #         $  f         g                        /         0  r         s                        ;         <  ~                                 G         H                          "         "  S#         T#  $         $  %         %  '         '  [(         `(  )         )  *         *  $,         ),  g-         l-  .         .  /         /  01         51  s2         x2  3         3  4         4  <6         A6  7         7  8         8  :         
:  H;         M;  <         <  =         =  ?         ?  T@         Y@  A         A  B         B  D         "D  `E         eE  F         i  j         j  7l         <l  zm         m  n         n   p         p  Cq         Hq  r         r  s         s  u         u  Ov         Tv  w         w  x         x  z         z  [{         `{  |         |  }         }  $         )  g         l                        0         5  s         x                        <         A                      ǋ           
  H         M             ΐ         Ӑ             T         Y                        ؕ  ؖ                      P  P                               >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               >!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     =
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,-y-     y--    -6-    6-E-    E-mx-l    mx-8-    8--x    -ib-	    ib-W-
    W-e-
    e--|    -ֳ-    ֳ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               1!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     6            Q  )          l  D            ~                                  8	  /
          s
  j                                 L         M                                   G         H  |         }                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                     0  /          s  r                                 zI  J         J  K         K  3M         4M  qN         rN  O         O  P         P  +R         ,R  iS         jS  T         T  U         U  #W         $W  aX         bX  Y         Y  Z         Z  \         \  Y]         Z]  ^         ^  _         _  a         a  Qb         Rb  c         c  d         d  f         f  Ig         Jg  h         h  i         i  k         k  @l         Bl  ~m         m  n         n  o         o  8q         :q  vr         xr  s         s  t         t  0v         2v  nw         pw  x         x  y         y  ({         *{  f|         h|  }         }  ~         ~            "  ^         `             ڃ         ܃             V         X             ҈         Ԉ             N         P             ʍ         ̍           
  F         H                      Ē              >         @  |         ~                        6         8  t         v                        .         0  l         n                        &         (  d         f                                    \         ^             ج         ڬ             T         V             б         ұ             M         N             ɶ         ޶             X                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               0!         @!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    |
         }
             v      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    |
         }
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     u
            0     -,-X     ,--     -)-    )-5-x    5-qD-|    qD-w-\    w-7 -    7 --h    -{b-    {b-!n-	    !n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               :!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         %            J  3          o  X            }                                              (	  
          M
  4         r                                   E         F  z         4  h         i                                   <                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            A         C                        
           M         O                        
         
  Y         [                        "         $  e         g                        .         0  q         s                        :         <  }                                 F         H                          "         "  R#         T#  $         $  %         %  '         '  \(         `(  )         )  *         *  %,         ),  h-         l-  .         .  /         /  11         51  t2         x2  3         3  4         4  =6         A6  7         7  8         8  :         
:  I;         M;  <         <  =         =  ?         ?  U@         Y@  A         A  B         B  D         "D  aE         eE  F         i  j         j  8l         <l  {m         m  n         n  p         p  Dq         Hq  r         r  s         s  u         u  Pv         Tv  w         w  x         x  z         z  \{         `{  |         |  }         }  %         )  h         l                        1         5  t         x                        =         A             Ë         ǋ           
  I         M             ϐ         Ӑ             U         Y            Ț  ț                      @  @          |  |                               >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               :!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     >
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-;-x    ;-I-|    I-|-\    |-'9-    '9--h    -b-    b-n-	    n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               6!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             4         5  i         j                                  r  R            w                                              +            G            c  ;            W            s         p                                     C!         D!  x"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            :         <  v         x                        *         ,  f         h             	         	             V         X                        
           F         H                                   6         8  r         t                        &         (  b         d                                     R!         T!  "         "  #         #  %         %  C&         D&  '         '  (         (  )         )  3+         4+  o,         p,  -         -  .         .  #0         $0  _1         `1  2         2  3         3  5         5  O6         P6  7         7  8         8  :         :  ?;         @;  {<         |<  =         =  >         >  /@         0@  kA         lA  B         B  C         C  E          E  [F         \F  G          _  n`          `  a          a  b          c  d          Hd  6e         m  o         o  Op           ֕         ؕ             N         P             ƚ         Ț             >         @  z         |                        .         0  j         l                                    Z         \             Ҫ         Ԫ             J         L             ¯         į              :         <  v         x                        *         ,  f         h                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               9!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     +
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -;-    ;-;-x    ;-}J-|    }J-}-\    }-]6-    ]6--h    -ib-    ib-}x-	    }x--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               B!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 9	         :	  n
         o
                         4         5  i         j                                               J  6            q                                              3   !          O!  '"          k"  C#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            B         C                                   N         O                        
         
  Z         [                        #         $  f         g                        /         0  r         s                        ;         <  ~                                 G         H                          "         "  S#         T#  $         $  %         F  G         G  -I         .I  pJ         qJ  K         K  L         L  9N         :N  |O         }O  P         P  R         R  ES         FS  T         T  U         U  W         W  QX         RX  Y         Y  Z         Z  \         \  ]]         ^]  ^         ^  _         _  &a         'a  ib         jb  c         c  d         d  2f         3f  ug         vg  h         h  i         i  >k         ?k  l         l  m         m  o         o  Fp         Kp  q         q  r         r  t         t  Ru         Wu  v         v  w         w  y          y  ^z         cz  {         {  |         |  '~         ,~  j         o                        3         8  v         {                        ?         D             Ŋ         ʊ             K          0            `  D            t                        Ԟ         ұ             S         X             ٶ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               /!         >!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 o
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-;-x    ;-I-|    I-|-\    |-'9-    '9--h    -b-    b-n-	    n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               8!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             4         5  i         j                                  r  Q            v                                              +            F            a  9          |  T            o         k                        	          
   >!         ?!  s"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ;         <  w         x                        +         ,  g         h             	         	             W         X                                   G         H                                   7         8  s         t                        '         (  c         d                                     S!         T!  "         "  #         #  %         %  B&         D&  ~'         '  (         (  )         )  2+         4+  n,         p,  -         -  .         .  "0         $0  ^1         `1  2         2  3         3  5         5  N6         P6  7         7  8         8  :         :  >;         @;  z<         |<  =         =  >         >  .@         0@  jA         lA  B         B  C         C  E          E  ZF         \F  G          ze  hf          f  g          g  h          i  i          Bj  0k         m  o         o  Np           ו         ؕ             O         P             ǚ         Ț             ?         @  {         |                        /         0  k         l                                    [         \             Ӫ         Ԫ             K         L             ï         į              ;         <  w         x                        +         ,  g         h                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               9!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     *
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-;-x    ;-}J-|    }J-}-\    }-]6-    ]6--h    -ob-    ob-}x-	    }x--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               >!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      	         
  >	         ?	  s
         t
                        9         :  n         o                                               O  <            w                                               6   !          Q!  )"          l"  D#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            A         C                        
           M         O                        
         
  Y         [                        "         $  e         g                        .         0  q         s                        :         <  }                                 F         H                          "         "  R#         T#  $         $  %         F  G         G  ,I         .I  oJ         qJ  K         K  L         L  8N         :N  {O         }O  P         P  R         R  DS         FS  T         T  U         U  W         W  PX         RX  Y         Y  Z         Z  \         \  \]         ^]  ^         ^  _         _  %a         'a  hb         jb  c         c  d         d  1f         3f  tg         vg  h         h  i         i  =k         ?k  l         l  m         m  o         o  Gp         Kp  q         q  r         r  t         t  Su         Wu  v         v  w         w  y          y  _z         cz  {         {  |         |  (~         ,~  k         o                        4         8  w         {                        @         D             Ɗ         ʊ  	           L                       P  4            d                        Ĥ         ұ             T         X             ڶ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               1!         B!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 o
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-6-x    6-+E-|    +E-yx-\    yx--    --h    -b-    b-n-	    n--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               8!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     8            T  ,          p  H                                              =	  8
          x
  s                                 Q         R                                   L         M                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            8         :  r         t                                  "  Z         \             	         	           
  B         D  |         ~                        *         ,  d         f                                   L         N                                   4         6  n         p                        !         !  V"         X"  #         #  $         H  J         J  WK         XK  L         L  M         M  O         O  ?P         @P  yQ         zQ  R         R  S         S  'U         (U  aV         bV  W         W  X         X  Z         Z  I[         J[  \         \  ]         ]  ^         ^  1`         2`  ka         la  b         b  c         c  e         e  Sf         Tf  g         g  h         h  j         j  ;k         <k  ul         vl  m          (n  
o         No  p         p  q         q  r         r  5t         6t  ou         pu  v         v  w         w  y         y  Wz         Xz  {         {  |         |  ~         ~  ?         @  y         z                        '         (  a         b             Շ         և             I         J                                   1         2  k         l             ߒ                                 @  "          f  H            n                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               <!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     +
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-5-x    5-}D-|    }D-w-\    w--    --h    -]b-    ]b-x-	    x--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               6!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         %  
          J  /          o  T            y                                              (	  	
          M
  .         r                                   E         F  z         4  h         i                                   <                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         8  *          n  `                                             %  6'         7'  y(         z(  )         )  *          +  B,         C,  -         -  .         .  0         0  N1         O1  2         2  3         3  5         5  Z6         [6  7         7  8         8  #:         $:  f;         g;  <         <  =         =  /?         0?  r@         s@  A         A  B         B  ;D         <D  ~E         E  F         F  H         H  GI         HI  J         J  K         K  M         M  ON         TN  O         O  P         P  R         R  [S         `S  T         T  U         U  $W         )W  gX         lX  Y         Y  Z         Z  0\         5\  s]         x]  ^         ^  _         _  <a         Aa  b         b  c         c  e         
e  Hf         Mf  g         g  h         h  j         j  Tk         Yk  l         l  m         m  o         "o  `p         ep  q         q  r         r  )t         .t  lu         qu  v         v  w         w  5y         :y  xz         }z  {         {  |         }  A~         F~             ǀ         ̀  
           M         R             Ӆ         ؅             Y         ^             ߊ           "         '  e         j                        .         3  q         v                        :         ?  }                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               (!         0!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 o
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-6-x    6-%E-|    %E-sx-\    sx--    --h    -b-    b-n-	    n--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               :!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     6            Q  )          l  D                                              8	  2
          s
  m                                 L         M                                   G         H  |         }                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            9         :  s         t                        !         "  [         \             	         	  	         
  C         D  }         ~                        +         ,  e         f                                   M         N                                   5         6  o         p                        !         !  W"         X"  #         #  $         H  J         J  VK         XK  L         L  M         M  O         O  >P         @P  xQ         zQ  R         R  S         S  &U         (U  `V         bV  W         W  X         X  Z         Z  H[         J[  \         \  ]         ]  ^         ^  0`         2`  ja         la  b         b  c         c  e         e  Rf         Tf  g         g  h         h   j         j  :k         <k  tl         vl  m         No  p         p  q         q  r         r  4t         6t  nu         pu  v         v  w         w  y         y  Vz         Xz  {         {  |         |  ~         ~  >         @  x         z                        &         (  `         b             ԇ         և             H         J                                   0         2  j         l             ޒ                                 ؙ                        $            J  ,                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               :!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     *
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -] -X     ] -հ-     հ-%ܰ-    %ܰ-0-p    0-;-T    ;--d-4    -d--    --8    -cb-    cb-x-	    x--8
    -E׳-
                   =         b      J      7   U   7   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               2!         >!  51                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 64 })  outputElements  5184    input0Elements  31104   input1Elements  24576   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     4                           #         b      J      7   M     operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       @                         %  	          J  .          o  S            v                                             (	  \
         ]
                                                         .         /  c                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     4                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  1728    input0Elements  1728    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       x         F  8          |  n              !          !  "         %  5'         7'  x(         z(  )         )  *          +  A,         C,  -         -  .         .  
0         0  M1         O1  2         2  3         3  5         5  Y6         [6  7         7  8         8  ":         $:  e;         g;  <         <  =         =  .?         0?  q@         s@  A         A  B         B  :D         <D  }E         E  F         F  H         H  FI         HI  J         J  K         K  M         M  PN         TN  O         O  P         P  R         R  \S         `S  T         T  U         U  %W         )W  hX         lX  Y         Y  Z         Z  1\         5\  t]         x]  ^         ^  _         _  =a         Aa  b         b  c         c  e         
e  If         Mf  g         g  h         h  j         j  Uk         Yk  l         l  m         m  o         "o  ap         ep  q         q  r         r  *t         .t  mu         qu  v         v  w         w  6y         :y  yz         }z  {         {  |         }  B~         F~             Ȁ         ̀             N         R             ԅ         ؅             Z         ^                        #         '  f         j                        /         3  r         v                        ;         ?  ~                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               *!         4!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 o
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -S-    S-;-x    ;-J-|    J-k}-\    k}-M-    M--h    -b-    b-qx-	    qx--h
    -K׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               O!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      "         #  W	         X	  
         
           {                                   N         O              \                                              4            `  8            `                          !           "  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            7         8  o         p                                   O         P             	         	  
         
  /         0  g         h                                   G         H                                   '         (  _         `                                   ?         @  w         x                          "          "  W#         X#  $         $  %         %  &          '  6(         8(  n)         p)  *         *  +         +  -         -  N.         P.  /         /  0         0  1         1  .3         03  f4         h4  5         5  6         6  8         8  F9         H9  ~:         :  ;         ;  <         <  &>         (>  ^?         `?  @         @  A         A  C         C  >D         @D  vE         xE  F         F  G         G  I         I  K         K  ML         NL  M         M  N         N  O         O  -Q         .Q  eR         fR  S         S  T         T  V         V  EW         FW  }X         ~X  Y         Y  Z         Z  %\         &\  ]]         ^]  ^         ^  _         _  a         a  =b         >b  uc         vc  d         d  e         e  g         g  Uh         Vh  i         i  j         j  k         k  5m         6m  mn          {  |          }  }          F~  5            x          ̀                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               8!         >!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 ^
  ~      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-<-x    <-J-|    J-}-\    }-w-    w--P    -{b-    {b-c-	    c-=-D
    =-K׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               :!         <!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 H	         I	  }
         ~
                        C         D  x         y                                     	          Y  D                                  
            &            B   !          ^!  6"          z"  R#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            5         6  k         l                                   C         D  y         z  	         	  
         
             Q         R                                   )         *  _         `                                   7         8  m         n                                   E         F  {         |               !         !  #         #  S$         T$  %         %  &         &  '         '  +)         ,)  a*         b*  +         +  ,         ,  .         .  9/         :/  o0         p0  1         1  2         2  4         4  G5         H5  }6         ~6  7         7  8         8  :          :  U;         V;  <         <  =         =  >         >  -@         .@  cA         dA  B         B  C         C  E         E  ;F         <F  pG         rG  H         H  I          I  J          K  
L          NL  BM          M  zN          N  O         nn  o         o  p         p  r         r  Ds         Fs  zt         |t  u         u  v         v  x         x  Ry         Ty  z         z  {         {  |         |  *~         ,~  `         b             ́         ΁             8         :  n         p             ڇ         ܇             F         H  |         ~                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               2!         @!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 _
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -M-    M-;-x    ;-J-|    J-k}-\    k}--    --h    -b-    b-qx-	    qx--h
    -K׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               K!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      "         #  W	         X	  
         
           {                                   N         O              \                                              0            [  3            \                          !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            6         8  n         p                                   N         P             	         	  
         
  .         0  f         h                                   F         H  ~                                 &         (  ^         `                                   >         @  v         x                          "          "  V#         X#  $         $  %         %  &          '  7(         8(  o)         p)  *         *  +         +  -         -  O.         P.  /         /  0         0  1         1  /3         03  g4         h4  5         5  6         6  8         8  G9         H9  :         :  ;         ;  <         <  '>         (>  _?         `?  @         @  A         A  C         C  ?D         @D  wE         xE  F         F  G         G  I         I  K         K  LL         NL  M         M  N         N  O         O  ,Q         .Q  dR         fR  S         S  T         T  V         V  DW         FW  |X         ~X  Y         Y  Z         Z  $\         &\  \]         ^]  ^         ^  _         _  a         a  <b         >b  tc         vc  d         d  e         e  g         g  Th         Vh  i         i  j         j  k         k  4m         6m  ln                      R  A                      ؅  ǆ            
                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               :!         B!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 _
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-;-x    ;-J-|    J-}-\    }-}-    }--P    -b-    b-c-	    c-=-D
    =-K׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               9!         :!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 C	         D	  x
         y
             	         
  >         ?  s         t                                               T  >            y                                               ;   !          V!  ."          q"  I#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            4         6  j         l                                   B         D  x         z  	         	  
         
             P         R                                   (         *  ^         `                                    6         8  l         n                                   D         F  z         |               !         !  #         #  R$         T$  %         %  &         &  '         '  *)         ,)  `*         b*  +         +  ,         ,  .         .  8/         :/  n0         p0  1         1  2         2  4         4  F5         H5  |6         ~6  7         7  8         8  :          :  T;         V;  <         <  =         =  >         >  ,@         .@  bA         dA  B         B  C         C  E         E  :F         <F  qG         rG  H         H  I          O  P          .Q  "R          fR  ZS          S  T          T  U         nn  o         o  p         p  r         r  Es         Fs  {t         |t  u         u  v         v  x         x  Sy         Ty  z         z  {         {  |         |  +~         ,~  a         b             ́         ΁             9         :  o         p             ۇ         ܇             G         H  }         ~                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               4!         @!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 _
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-6-x    6-1E-|    1E-x-\    x-7-    7--h    -cb-    cb-ex-	    ex--h
    -K׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               A!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     8            T  ,          p  H            z                                  =	  +
          x
  f                                 Q         R                                   L         M                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         <  2            u                                  H  >         U$  %         %  &         &  '         '  ,)         -)  b*         c*  +         +  ,         ,  .         .  :/         ;/  p0         q0  1         1  2         2  4         4  H5         I5  ~6         6  7         7  8         8   :         !:  V;         W;  <         <  =         =  >         >  .@         /@  dA         eA  B         B  C         C  E         E  <F         =F  rG         G  I         I  CJ         EJ  yK         {K  L         L  M         M  O         O  QP         SP  Q         Q  R         R  S         S  )U         +U  _V         aV  W         W  X         X  Z         Z  7[         9[  m\         o\  ]         ]  ^         ^  `         `  Ea         Ga  {b         }b  c         c  d         d  f         f  Sg         Ug  h         h  i         i  j         j  +l         l  
n         ړ             E         F  {         |                                   S         T                                   +         ,  a         b             ͤ         Τ             9         :  o         p             ۪         ܪ             G         H  }         ~                                    U         V                      µ             -                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               /!         0!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 [
  {      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -5-    5-#;-    #;-I-    I-}-d    }--    --p    -b-    b-c-
    c-q-
    q-۰-t    ۰-K׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               A!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             4         5  i         j                                                                                  2  
          N  0          s  U            z                                 p                                     C!         D!  x"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                 $  #          g  f                     T#  $         $  %         %  &         &  3(         4(  k)         l)  *         *  +         +  -         -  K.         L.  /         /  0         0  1         1  +3         ,3  c4         d4  5         5  6         6  8         8  C9         D9  {:         |:  ;         ;  <         <  #>         $>  [?         \?  @         @  A         A  C         C  ;D         <D  sE         tE  F         F  G         G  I         I  RJ         TJ  K         K  L         L  M         M  2O         4O  jP         lP  Q         Q  R         R  T         T  JU         LU  V         V  W         W  X         X  *Z         ,Z  b[         d[  \         \  ]         ]  
_         _  B`         D`  za         |a  b         b  c         c  "e         $e  Zf         \f  g         g  h         h  j         j  :k         <k  rl         tl  m         m  n         n  p         p  Rq         Tq  r         r  s         s  t         t  2v         4v  jw         lw  x         x  y         y  {         {  J|         L|  }         }  ~         ~             *         ,  b         d             ҄         Ԅ  
           B         D  z         |                        "         $  Z         \             ʏ         ̏                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               =!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    |
         }
             v      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     &
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -o -X     o -ְ-     ְ-7ܰ-    7ܰ-0-p    0-U<-T    U<-d-4    d-7-    7--8    -ib-    ib-ex-	    ex--8
    -K׳-
                   =         b      J      7   U   7   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               =!         >!  @1                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 64 })  outputElements  5184    input0Elements  31104   input1Elements  24576   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     4                           #         b      J      7   M     operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       @                                     6            Q  )          l  [                                    	         X	  
         
                        +           L         M                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     4                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  1728    input0Elements  1728    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       x                         !         U$  %         %  &         &  '         '  +)         -)  a*         c*  +         +  ,         ,  .         .  9/         ;/  o0         q0  1         1  2         2  4         4  G5         I5  }6         6  7         7  8         8  :         !:  U;         W;  <         <  =         =  >         >  -@         /@  cA         eA  B         B  C         C  E         E  ;F         =F  qG         G  I         I  DJ         EJ  zK         {K  L         L  M         M  O         O  RP         SP  Q         Q  R         R  S         S  *U         +U  `V         aV  W         W  X         X  Z         Z  8[         9[  n\         o\  ]         ]  ^         ^  `         `  Fa         Ga  |b         }b  c         c  d         d  f         f  Tg         Ug  h         h  i         i  j         j  ,l         l  	n          n  o          Zo  Pp         ړ             D         F  z         |                                   R         T                                   *         ,  `         b             ̤         Τ             8         :  n         p             ڪ         ܪ             F         H  |         ~                                    T         V                      µ             ,                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               1!         4!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 [
  {      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -;-    ;-;-    ;-I-    I-}-d    }--    --p    -b-    b-c-
    c-q-
    q-۰-t    ۰-K׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               E!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             4         5  i         j                                                                                  .            I  *          n  O            t                                 k                        	          
   >!         ?!  s"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                     0  /          s  r                                 T#  $         $  %         %  &         &  2(         4(  j)         l)  *         *  +         +  -         -  J.         L.  /         /  0         0  1         1  *3         ,3  b4         d4  5         5  6         6  
8         8  B9         D9  z:         |:  ;         ;  <         <  ">         $>  Z?         \?  @         @  A         A  C         C  :D         <D  rE         tE  F         F  G         G  I         I  SJ         TJ  K         K  L         L  M         M  3O         4O  kP         lP  Q         Q  R         R  T         T  KU         LU  V         V  W         W  X         X  +Z         ,Z  c[         d[  \         \  ]         ]  _         _  C`         D`  {a         |a  b         b  c         c  #e         $e  [f         \f  g         g  h         h  j         j  ;k         <k  sl         tl  m         m  n         n  p         p  Sq         Tq  r         r  s         s  t         t  3v         4v  kw         lw  x         x  y         y  {         {  K|         L|  }         }  ~         ~             +         ,  c         d             ӄ         Ԅ             C         D  {         |                        #         $  [         \             ˏ         ̏                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               =!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    |
         }
             v      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     %
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-_;-x    _;-I-|    I-G}-\    G}-/-    /--P    -ub-    ub-c-	    c-I-D
    I-E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               8!         <!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      "         #  W	         X	  
         
           {                                   N         O              h                                                          =  %          b  J            o              !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            >         C                                   J         O                        
         
  V         [                                 $  b         g                        +         0  n         s                        7         <  z                                  C         H                          "          "  "          3#  $         2  3         3  	5         5  L6         Q6  7         7  8         8  :         :  X;         ];  <         <  =         =  !?         &?  d@         i@  A         A  B         B  -D         2D  pE         uE  F         F  G         G  9I         >I  |J         J  K         K  M         M  EN         JN  O         O  P         P  R         R  QS         VS  T         T  U         U  W          oi  Pj          j  rk          k  l         l  n         n  Uo         Zo  p         p  q         q   s         !s  bt         ct  u         u  v         v  *x         +x  my         ny  z         z  {         {  6}         7}  y~         z~                         B         C             Ȅ         Ʉ             N         O             ԉ         Չ             Z         [                        #         v                        >                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               0!         @!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 k
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    <     -,-p     ,--     -5-    5-5-    5-}D-    }D-w-t    w-U2-     U2-9-X    9--    -b-p	    b-]-x
    ]-k-
    k--    -ֳ-\    ֳ--<                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ?!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         %            J  ,          o  Q            v                                              (	  
          M
  +         r                                   E         F  z         4  h         i                                   <                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        "  #         #  !%         #%  [&         ]&  '         '  (         (  	*         *  C+         E+  },         ,  -         -  .         .  +0         -0  e1         g1  2         2  3         3  5         5  M6         O6  7         7  8         8  9         9  5;         7;  o<         q<  =         =  >         >  @         @  WA         YA  B         B  C         C  E         E  ?F         h  8i         :i  rj         tj  k         k  l         l   n         "n  Zo         \o  p         p  q         q  s         
s  Bt         Dt  |u         ~u  v         v  w         w  *y         ,y  dz         fz  {         {  |         |  ~         ~  L         N                                   4         6  n         p                                   V         X             ʌ         ̌             =         ?  w         y                        #         %  ]         _             ї         ӗ             E         G                                   -         /  g         i             ۢ         ݢ             O         Q                        ŧ                        9  2          s  l                               !  Y         [             Ͳ                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               >!         H!  1         1  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    |
         }
             v                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    |
         }
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     e
            0     -,-X     ,--     --    -_;-x    _;-I-|    I-G}-\    G}-/-    /--P    -ub-    ub-c-	    c-I-D
    I-E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               5!         :!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      "         #  W	         X	  
         
           {                                   N         O              i                                                          =  $          b  I            n              !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            ?         C                                   K         O                        
         
  W         [                                  $  c         g                        ,         0  o         s                        8         <  {                                 D         H                          "         2  3         3  
5         5  M6         Q6  7         7  8         8  :         :  Y;         ];  <         <  =         =  "?         &?  e@         i@  A         A  B         B  .D         2D  qE         uE  F         F  G         G  :I         >I  }J         J  K         K  M         M  FN         JN  O         O  P         P  R         R  RS         VS  T         T  U         U  W         l  n         n  Vo         Zo  p         p  q         q  s         !s  at         ct  u         u  v         v  )x         +x  ly         ny  z         z  {         {  5}         7}  x~         z~                         A         C             Ǆ         Ʉ  
           M         O             Ӊ         Չ             Y         [             ߎ           "          $            F  '          h  I            k                     v                        =                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               0!         @!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 k
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    <     -,-p     ,--     -;-    ;-5-    5-}D-    }D-w-t    w-2-     2-9-X    9--    -b-p	    b-]-x
    ]-k-
    k--    -ֳ-\    ֳ--<                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               C!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         %            J  +          o  P            u                                              (	  
          M
  ,         r                                   E         F  z         4  h         i                                   <                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        "  #         #  "%         #%  \&         ]&  '         '  (         (  
*         *  D+         E+  ~,         ,  -         -  .         .  ,0         -0  f1         g1  2         2  3         3  5         5  N6         O6  7         7  8         8  9         9  6;         7;  p<         q<  =         =  >         >  @         @  XA         YA  B         B  C         C  E         E  @F         h  9i         :i  sj         tj  k         k  l         l  !n         "n  [o         \o  p         p  q         q  	s         
s  Ct         Dt  }u         ~u  v         v  w         w  +y         ,y  ez         fz  {         {  |         |  ~         ~  M         N                                   5         6  o         p                                   W         X             ˌ         ̌             >         ?  x         y                        $         %  ^         _             җ         ӗ             F         G                                   .         /  h         i             ܢ         ݢ             P         Q             ħ         ŧ             8         9  r         s                                    !            [  T                                          b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               @!         H!  1         1  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    |
         }
             v                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    |
         }
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     f
            4     -,-`     ,--     -)-    )-;-    ;-I-    I-|-d    |--    --p    -b-    b-]-
    ]-k-
    k--t    -?׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ?!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             4         5  i         j                                  r  Z                                                          +            G            c  ;            W            s         p                                     C!         D!  x"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                     &            X  I            {                     %  F&         K&  '         '  (         (  *         *  R+         W+  ,         ,  -         -  /          /  ^0         c0  1         1  2         2  '4         ,4  j5         o5  6         6  7         7  39         89  v:         {:  ;         ;  <         =  ?>         D>  ?         ?  @         @  B         B  KC         PC  D         D  E         E  G         G  WH         \H  I         I  J         J   L         %L  cM         hM  N         N  O         O  ,Q         1Q  oR         tR  S         S  T         T  8V         =V  {W         W  X         X  Z         Z  D[         I[  \         \  ]         ]  _         _  P`         U`  a         a  b         b  d         d  \e         ae  f         f  g         g  %i         *i  hj         mj  k         k  l         l  5n         6n  xo         yo  p         p  q         q  As         Bs  t         t  u         u  
w         w  Mx         Nx  y         y  z         z  |         |  Y}         Z}  ~         ~             "         #  e         f                        .         /  q         r                        :         ;  }         ~                        F         G             ̒         ͒             N         S                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ;!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    |
         }
             v      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     B
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-;-x    ;-I-|    I-|-\    |--    --h    -b-    b-!n-	    !n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               :!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             4         5  i         j                                                                                  2  
          N  0          s  U            z                                 p                                     C!         D!  x"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ;         <  w         x                        +         ,  g         h             	         	             W         X                        
           F         H                                   6         8  r         t                        &         (  b         d                                     R!         T!  "         "  #         #  %          %  &          &  '          (  (          $)  )          B*  +         :H  uI         vI  J         J  K         K  )M         *M  eN         fN  O         O  P         P  R         R  US         VS  T         T  U         U  	W         
W  EX         FX  Y         Y  Z         Z  [         [  5]         6]  q^         r^  _         _  `         `  %b         &b  ac         bc  d         d  e         e  g         g  Qh         Rh  i         i  j         j  l         l  Am         Bm  |n         ~n  o         o  p         p  0r         2r  ls         ns  t         t  u         u   w         "w  \x         ^x  y         y  z         z  |         |  L}         N}  ~         ~                         <         >  x         z                        ,         .  h         j                                   X         Z             Џ         ҏ             H                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ;!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     =
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -/-    /-;-    ;-I-    I-|-d    |--    --p    -b-    b-]-
    ]-k-
    k--t    -?׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               A!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             4         5  i         j                                  r  Y            ~                                              +            F            a  9          |  T            o         k                        	          
   >!         ?!  s"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                  R  C            u                     %  G&         K&  '         '  (         (  *         *  S+         W+  ,         ,  -         -  /          /  _0         c0  1         1  2         2  (4         ,4  k5         o5  6         6  7         7  49         89  w:         {:  ;         ;  <         =  @>         D>  ?         ?  @         @  	B         B  LC         PC  D         D  E         E  G         G  XH         \H  I         I  J         J  !L         %L  dM         hM  N         N  O         O  -Q         1Q  pR         tR  S         S  T         T  9V         =V  |W         W  X         X  Z         Z  E[         I[  \         \  ]         ]  _         _  Q`         U`  a         a  b         b  d         d  ]e         ae  f         f  g         g  &i         *i  ij         mj  k         k  l         l  4n         6n  wo         yo  p         p  q         q  @s         Bs  t         t  u         u  	w         w  Lx         Nx  y         y  z         z  |         |  X}         Z}  ~         ~             !         #  d         f                        -         /  p         r                        9         ;  |         ~                        E         G             ˒         ͒             O         S                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               9!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    |
         }
             v      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     A
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-;-x    ;-I-|    I-|-\    |--    --h    -b-    b-!n-	    !n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               <!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             4         5  i         j                                                                                  .            I  *          n  O            t                                 k                        	          
   >!         ?!  s"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            :         <  v         x                        *         ,  f         h             	         	             V         X                                   G         H                                   7         8  s         t                        '         (  c         d                                     S!         T!  "         "  #         #  %          `+  :,          ~,  X-          -  v.          .  /          /  0         :H  tI         vI  J         J  K         K  (M         *M  dN         fN  O         O  P         P  R         R  TS         VS  T         T  U         U  W         
W  DX         FX  Y         Y  Z         Z  [         [  4]         6]  p^         r^  _         _  `         `  $b         &b  `c         bc  d         d  e         e  g         g  Ph         Rh  i         i  j         j  l         l  @m         Bm  }n         ~n  o         o  p         p  1r         2r  ms         ns  t         t  u         u  !w         "w  ]x         ^x  y         y  z         z  |         |  M}         N}  ~         ~                        =         >  y         z                        -         .  i         j                                   Y         Z             я         ҏ             I                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ?!         R!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     >
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    <     -,-p     ,--     -)-    )-6-    6-%E-    %E-sx-t    sx-u9-     u9-A-X    A--    -b-p	    b-W-x
    W-e-
    e--    -ֳ-\    ֳ--<                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               =!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     8            T  ,          p  H            z                                  =	  +
          x
  f                                 Q         R                                   L         M                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        D%  &         I  J         J  4L         5L  wM         xM  N         N  O         O  @Q         AQ  R         R  S         S  	U         
U  LV         MV  W         W  X         X  Z         Z  X[         Y[  \         \  ]         ]  !_         "_  d`         e`  a         a  b         b  -d         .d  pe         qe  f         f  g         g  9i         :i  |j         }j  k         k  m         m  An         Fn  o         o  p         p  r         r  Qs         Rs  t         t  u         u  w         w  ]x         ^x  y         y  z         z  &|         '|  i}         j}  ~         ~             2         3  u         v                        >         ?             Ĉ         ň             J         K             Ѝ         э             V         W             ܒ         ݒ              ^         c                        +         ,  j         o                        3         8  v         {                        ?         D             ť         ʥ             K         P             Ѫ         ֪             W          \  [                                  %  $          h  g                      ,         1  o         t                                           b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               <!         H!  1         1  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    |
         }
             v                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    |
         }
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            0     -,-X     ,--     --    -5-x    5-qD-|    qD-w-\    w-q-    q--h    -ob-    ob-n-	    n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               6!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         %            J  ,          o  Q            v                                              (	  
          M
  +         r                                   E         F  z         4  h         i                                   <                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            =         >  {         |                        5         6  s         t             	         	  -         .  k         l                        %         &  c         d                                   [         \                                   S         T                                   K          L   !         !  "         "  $         $  C%          L1  62          z2  d3          3  4          4  5          6  6         G  H         H  0J         2J  nK         pK  L         L  M         M  (O         *O  fP         hP  Q         Q  R         R   T         "T  ^U         `U  V         V  W         W  Y         Y  VZ         XZ  [         [  \         \  ^         ^  N_         P_  `         `  a         a  c         
c  Fd         Hd  e         e  f         f   h         h  >i         @i  |j         ~j  k         k  l         n  o         o  *q         ,q  hr         jr  s         s  t         t  "v         $v  `w         bw  x         x  y         y  {         {  X|         Z|  }         }  ~         ~             P         R             ̃         ΃  
           H         J             Ĉ         ƈ             @         B  ~                                 8         :  v         x                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               2!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     =
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    <     -,-p     ,--     -/-    /-6-    6-E-    E-mx-t    mx-99-     99-@-X    @--    -b-p	    b-W-x
    W-e-
    e--    -ֳ-\    ֳ--<                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ?!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     6            Q  )          l  D            v                                  8	  '
          s
  b                                 L         M                                   G         H  |         }                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        D%  &         I  J         J  3L         5L  vM         xM  N         N  O         O  ?Q         AQ  R         R  S         S  U         
U  KV         MV  W         W  X         X  Z         Z  W[         Y[  \         \  ]         ]   _         "_  c`         e`  a         a  b         b  ,d         .d  oe         qe  f         f  g         g  8i         :i  {j         }j  k         k  m         m  Bn         Fn  o         o  p         p  r         r  Ps         Rs  t         t  u         u  w         w  \x         ^x  y         y  z         z  %|         '|  h}         j}  ~         ~             1         3  t         v                        =         ?             È         ň             I         K             ύ         э             U         W             ے         ݒ              _         c                        *         ,  k         o                        4         8  w         {                        @         D             ƥ         ʥ  	           L         P             Ҫ         ֪             X         \             ޯ           !         %  d         h                                    1  0          t  s                                          b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               <!         H!  1         1  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    |
         }
             v                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    |
         }
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            0     -,-X     ,--     -#-    #-5-x    5-qD-|    qD-w-\    w-w-    w--h    -ub-    ub-!n-	    !n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               8!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         %            J  +          o  P            u                                              (	  
          M
  ,         r                                   E         F  z         4  h         i                                   <                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            <         >  z         |                        4         6  r         t             	         	  ,         .  j         l                        $         &  b         d                                   Z         \                                   R         T                                   J          L   !         !  "         "  $         $  B%          27  8          `8  J9          9  x:          :  ;          ;  <         G  H         H  1J         2J  oK         pK  L         L  M         M  )O         *O  gP         hP  Q         Q  R         R  !T         "T  _U         `U  V         V  W         W  Y         Y  WZ         XZ  [         [  \         \  ^         ^  O_         P_  `         `  a         a  	c         
c  Gd         Hd  e         e  f         f  h         h  ?i         @i  }j         ~j  k         k  l         n  o         o  +q         ,q  ir         jr  s         s  t         t  #v         $v  aw         bw  x         x  y         y  {         {  Y|         Z|  }         }  ~         ~             Q         R             ̓         ΃             I         J             ň         ƈ             A         B                                   9         :  w         x                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               4!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     >
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-;-x    ;-I-|    I-|-\    |-7-    7--h    -b-    b-	n-	    	n--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               8!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             4         5  i         j                                  r  Z                                                          +            G            c  ;            W            s         p                                     C!         D!  x"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            >         @  ~                                  >         @  ~                      	          
  >         @  ~                                  >         @  ~                                  >         @  ~                                  >         @  ~                                  >         @  ~             !         !  "          #  >$         @$  ~%          %  &          &  '          '  (          &)  *          T*  >+         K  L         L  /N         0N  oO         pO  P         P  Q         Q  /S         0S  oT         pT  U         U  V         V  /X         0X  oY         pY  Z         Z  [         [  /]         0]  o^         p^  _         _  `         `  /b         0b  oc         pc  d         d  e         e  /g         0g  oh         ph  i         i  j         j  /l         0l  om         pm  n         n  o         o  /q         0q  or           2         4  r         t                        2         4  r         t                        2         4  r         t                        2         4  r         t                        2         4  r         t                        2         4  r         t                        2         4  r         t                        2                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ;!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     +
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     --    -<-x    <-J-|    J-}-\    }--    --h    -ub-    ub-wx-	    wx--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               4!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      ,         -  a	         b	  
         
             '         (  \         ]                                              7  +          r  f                                  #            ?             [   3!          w!  O"          "  k#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ?         @                                    ?         @                        	          
  ?         @                                    ?         @                                    ?         @                                    ?         @                                    ?         @               !         !  "          #  ?$         @$  %         %  &         &  '          (  ?)         @)  *         *  +         +  ,          -  ?.         @.  /         /  0         0  1          2  ?3         @3  4         4  5         5  6          7  ?8         @8  9         9  :         :  ;          <  ?=         @=  >         >  ?         ?  @          A  ?B         @B  C         C  D         D  E          F  ?G         @G  H         H  I         I  J          K  >L         @L  ~M         M  N         N  O          P  >Q         @Q  ~R         R  S         S  T          U  >V         @V  ~W         W  X         X  Y          Z  >[         @[  ~\         \  ]         ]  ^          _  >`         @`  ~a         a  b         b  c          d  >e         @e  ~f         f  g         g  h          i  >j         @j  ~k         k  l         l  m                      ƈ                        B  <            z            ^         `                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               7!         F!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 o
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-;-x    ;-I-|    I-|-\    |-7-    7--h    -b-    b-	n-	    	n--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               8!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             4         5  i         j                                  r  Y            ~                                              +            F            a  9          |  T            o         k                        	          
   >!         ?!  s"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ?         @                                    ?         @                        	          
  ?         @                                    ?         @                                    ?         @                                    ?         @                                    ?         @               !         !  "          #  ?$         @$  %          +  l,          ,  -          -  .          /  /          :0  $1         K  L         L  .N         0N  nO         pO  P         P  Q         Q  .S         0S  nT         pT  U         U  V         V  .X         0X  nY         pY  Z         Z  [         [  .]         0]  n^         p^  _         _  `         `  .b         0b  nc         pc  d         d  e         e  .g         0g  nh         ph  i         i  j         j  .l         0l  nm         pm  n         n  o         o  .q         0q  nr           3         4  s         t                        3         4  s         t                        3         4  s         t                        3         4  s         t                        3         4  s         t                        3         4  s         t                        3         4  s         t                        3                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ?!         R!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     *
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-y-     y--    -<-x    <-J-|    J-}-\    }--    --h    -{b-    {b-wx-	    wx--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               2!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      1         2  f	         g	  
         
             ,         -  a         b                                               <  1          w  l                                  (             C             ^   6!          y!  Q"          "  l#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            >         @  ~                                  >         @  ~                      	          
  >         @  ~                                  >         @  ~                                  >         @  ~                                  >         @  ~                                  >         @  ~             !         !  "          #  >$         @$  ~%         %  &         &  '          (  >)         @)  ~*         *  +         +  ,          -  >.         @.  ~/         /  0         0  1          2  >3         @3  ~4         4  5         5  6          7  >8         @8  ~9         9  :         :  ;          <  >=         @=  ~>         >  ?         ?  @          A  >B         @B  ~C         C  D         D  E          F  >G         @G  ~H         H  I         I  J          K  ?L         @L  M         M  N         N  O          P  ?Q         @Q  R         R  S         S  T          U  ?V         @V  W         W  X         X  Y          Z  ?[         @[  \         \  ]         ]  ^          _  ?`         @`  a         a  b         b  c          d  ?e         @e  f         f  g         g  h          i  ?j         @j  k         k  l         l  m                                  :  4          x  r                        _         `                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               9!         J!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 o
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-y-     y--    -6-x    6-E-|    E-gx-\    gx--    --h    -ib-    ib-n-	    n--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               4!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     8            T  ,          p  H            z                                  =	  +
          x
  f                                 Q         R                                   L         M                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            <         >  z         |                        4         6  r         t             	         	  ,         .  j         l                        $         &  b         d                                   Z         \                                   R         T                                   J          L   !         !  "         "  $         $  B%          J1  $2          n2  H3          3  l4          4  5          5  6         H  !J         "J  _K         `K  L         L  M         M  O         O  WP         XP  Q         Q  R         R  T         T  OU         PU  V         V  W         W  	Y         
Y  GZ         HZ  [         [  \         \  ^         ^  ?_         @_  }`         ~`  a         a  b         b  7d         8d  ue         ve  f         f  g         g  /i         0i  mj         nj  k         k  l         l  'n         o  q         q  Er         Fr  s         s  t         t  u          v  =w         >w  {x         |x  y         y  z         z  5|         6|  s}         t}  ~         ~             -         .  k         l                        %         &  c         d             ߉                      [         \             ׎         ؎             S         T             ϓ         Г                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               2!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     +
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-y-     y--    -5-x    5-kD-|    kD-w-\    w-s-    s--h    -Wb-    Wb-qx-	    qx--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               4!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         %            J  ,          o  Q            v                                              (	  
          M
  +         r                                   E         F  z         4  h         i                                   <                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                        2            d  O                                 p&  '         '  (         (  8*         9*  {+         |+  ,         ,  .         .  D/         E/  0         0  1         1  3         3  P4         Q4  5         5  6         6  8         8  \9         ]9  :         :  ;         ;  %=         &=  h>         i>  ?         ?  @         @  1B         2B  tC         uC  D         D  E         E  =G         >G  H         H  I         I  K         K  IL         JL  M         M  N         N  P         P  QQ         VQ  R         R  S         S  U         U  ]V         bV  W         W  X         X  &Z         +Z  i[         n[  \         \  ]         ]  2_         7_  u`         z`  a         a  b          c  >d         Cd  e         e  f         f  h         h  Ji         Oi  j         j  k         k  m         m  Vn         [n  o         o  p         p  r         $r  bs         gs  t         t  u         u  +w         0w  nx         sx  y         y  z         z  7|         <|  z}         }  ~         ~              C         H             Ƀ         ΃             O         T             Ո         ڈ             [         `                        $         )  g         l                        0         5  s         x                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               .!         8!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 o
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-y-     y--    -y6-x    y6-E-|    E-ax-\    ax--    --h    -ib-    ib-n-	    n--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               4!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     6            Q  )          l  D            v                                  8	  '
          s
  b                                 L         M                                   G         H  |         }                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            =         >  {         |                        5         6  s         t             	         	  -         .  k         l                        %         &  c         d                                   [         \                                   S         T                                   K          L   !         !  "         "  $         $  C%          6  7          "8  8          F9   :          j:  D;          ;  h<         H   J         "J  ^K         `K  L         L  M         M  O         O  VP         XP  Q         Q  R         R  T         T  NU         PU  V         V  W         W  Y         
Y  FZ         HZ  [         [  \         \   ^         ^  >_         @_  |`         ~`  a         a  b         b  6d         8d  te         ve  f         f  g         g  .i         0i  lj         nj  k         k  l         l  &n         o  q         q  Dr         Fr  s         s  t         t  u          v  <w         >w  zx         |x  y         y  z         z  4|         6|  r}         t}  ~         ~             ,         .  j         l                        $         &  b         d             މ                      Z         \             ֎         ؎             R         T             Γ         Г                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               4!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     *
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-s-     s--    -5-x    5-kD-|    kD-w-\    w-s-    s--h    -]b-    ]b-qx-	    qx--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               2!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         %            J  +          o  P            u                                              (	  
          M
  ,         r                                   E         F  z         4  h         i                                   <                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                     ,            ^  I	          	  {
          
           p&  '         '  (         (  7*         9*  z+         |+  ,         ,   .         .  C/         E/  0         0  1         1  3         3  O4         Q4  5         5  6         6  8         8  [9         ]9  :         :  ;         ;  $=         &=  g>         i>  ?         ?  @         @  0B         2B  sC         uC  D         D  E         E  <G         >G  H         H  I         I  K         K  HL         JL  M         M  N         N  P         P  RQ         VQ  R         R  S         S  U         U  ^V         bV  W         W  X         X  'Z         +Z  j[         n[  \         \  ]         ]  3_         7_  v`         z`  a         a  b          c  ?d         Cd  e         e  f         f  h         h  Ki         Oi  j         j  k         k  m         m  Wn         [n  o         o  p         p   r         $r  cs         gs  t         t  u         u  ,w         0w  ox         sx  y         y  z         z  8|         <|  {}         }  ~         ~             D         H             ʃ         ΃             P         T             ֈ         ڈ             \         `                        %         )  h         l                        1         5  t         x                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               0!         <!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 o
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-y-     y--    -M;-x    M;-I-|    I-5}-\    5}--    --h    -{b-    {b-Mx-	    Mx--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               4!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      "         #  W	         X	  
         
           {                                   N         O              h                                                          =  %          b  J            o              !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            :         <  v         x                        *         ,  f         h             	         	             V         X                        
           F         H                                   6         8  r         t                        &         (  b         d                                     R!         T!  "         "  #         #  %         %  B&         D&  ~'         '  (         (  )         )  2+         4+  n,         p,  -         -  .         .  "0         $0  ^1         `1  2         2  3         3  5         5  N6         P6  7         7  8         8  :         :  >;         @;  z<         |<  =         =  >         >  .@         0@  jA         lA  B         B  C         C  E          E  ZF         \F  G         G  H         H  J         J  WK         XK  L         L  M         M  O         O  GP         HP  Q         Q  R         R  S         S  7U         8U  sV         tV  W         W  X         X  'Z         (Z  c[         d[  \         \  ]         ]  _         _  S`         T`  a         a  b         b  d         d  Ce         De  f         f  g         g  h         h  3j         4j  ok         pk  l         l  m         m  #o                        ݊                      \  Y                               >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               E!         F!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 [
  {      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-a-     a--    -;-x    ;-J-|    J-}-\    }-7-    7--h    -ub-    ub--	    --h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               3!         4!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      ,         -  a	         b	  
         
             '         (  \         ]                                              7  +          r  f                                  #            ?             [   3!          w!  O"          "  k#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            B         C                                   N         O                        
         
  Z         [                        #         $  f         g                        /         0  r         s                        ;         <  ~                                 G         H                          "         "  S#         T#  $         $  %         %  '         '  _(         `(  )         )  *         *  (,         ),  k-         l-  .         .  /         /  41         51  w2         x2  3         3  4         4  @6         A6  7         7  8         8  	:         
:  L;         M;  <         <  =         =  ?         ?  X@         Y@  A         A  B         B  !D         "D  dE         eE  F         F  G         G  -I         .I  pJ         qJ  K          XW  >X          X  hY          Y  Z          Z  [           \  \         n  o         o  !q         &q  dr         	  G         L             ͙         ҙ             S         X             ٞ         ޞ           !  _         d                        (         -  k         p                        4         9  w         |                        @         E             Ʊ         ˱  	           L         Q             Ҷ         ׶                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               C!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-s-     s--    -G;-x    G;-I-|    I-/}-\    /}-C-    C--h    -b-    b-Mx-	    Mx--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               2!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      "         #  W	         X	  
         
           {                                   N         O              i                                                          =  $          b  I            n              !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ;         <  w         x                        +         ,  g         h             	         	             W         X                                   G         H                                   7         8  s         t                        '         (  c         d                                     S!         T!  "         "  #         #  %         %  C&         D&  '         '  (         (  )         )  3+         4+  o,         p,  -         -  .         .  #0         $0  _1         `1  2         2  3         3  5         5  O6         P6  7         7  8         8  :         :  ?;         @;  {<         |<  =         =  >         >  /@         0@  kA         lA  B         B  C         C  E          E  [F         \F  G         G  H         H  J         J  VK         XK  L         L  M         M  
O         O  FP         HP  Q         Q  R         R  S         S  6U         8U  rV         tV  W         W  X         X  &Z         (Z  b[         d[  \         \  ]         ]  _         _  R`         T`  a         a  b         b  d         d  Be         De  ~f         f  g         g  h         h  2j         4j  nk         pk  l         l  m         m  "o          ؎  Տ                      T  Q                      Г  ͔                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               I!         J!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 [
  {      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-[-     [--    -;-x    ;-J-|    J-}-\    }-7-    7--h    -ob-    ob--	    --h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               0!         2!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      1         2  f	         g	  
         
             ,         -  a         b                                               <  1          w  l                                  (             C             ^   6!          y!  Q"          "  l#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            A         C                        
           M         O                        
         
  Y         [                        "         $  e         g                        .         0  q         s                        :         <  }                                 F         H                          "         "  R#         T#  $         $  %         %  '         '  ^(         `(  )         )  *         *  ',         ),  j-         l-  .         .  /         /  31         51  v2         x2  3         3  4         4  ?6         A6  7         7  8         8  :         
:  K;         M;  <         <  =         =  ?         ?  W@         Y@  A         A  B         B   D         "D  cE         eE  F         F  G         G  ,I         .I  oJ         qJ  K          *]  ^          T^  :_          ~_  d`          `  a          a  b         n  o         o  "q         &q  er         	  H         L             Ι         ҙ             T         X             ڞ         ޞ           !  `         d                        )         -  l         p                        5         9  x         |                        A         E             Ǳ         ˱  
           M         Q             Ӷ         ׶                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ?!         @!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-C-     C--    -6-x    6-E-|    E-gx-\    gx--    --h    -b-    b-Gx-	    Gx--h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     R                       T!  1          1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     <            Z  2          x  P                                              A	  3
          z
  l                                !  W         X                                   P         Q                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                        4  $          h  X                                 &  '         '  )         ")  `*         e*  +         +  ,         ,  ).         ..  l/         q/  0         0  1         1  53         :3  x4         }4  5         5  6         7  A8         F8  9         9  :         :  
<         <  M=         R=  >         >  ?         ?  A         A  YB         ^B  C         C  D         D  "F         'F  eG         jG  H         H  I         I  .K         3K  qL         vL  M         M  N         N  >P         ?P  Q         Q  R         R  T         T  JU         KU  V         V  W         W  Y         Y  VZ         WZ  [         [  \         \  ^          ^  b_         c_  `         `  a         a  +c         ,c  nd         od  e         e  f         f  7h         8h  zi         {i  j         j   l         l  Cm         Dm  n         n  o         o  q         q  Or         Pr  s         s  t         t  v         v  [w         \w  x         x  y         y  ${         %{  g|         h|  }         }  ~         ~  0         1  s         t                        <         =                      Ç             H         I             Ό         ό             T         U             ڑ         ۑ             `         a                        )                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               W!         X!  2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 [
  {      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -{,-X     {,-1-     1--    -;-x    ;-I-|    I-|-\    |--    --h    -?b-    ?b--	    --h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     Q                       F!  1          1  j=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             6         7  m         n                                                                                  :            X  >          {  a                                             f                        
             A!         B!  x"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            6         8  n         p                                   N         P             	         	  
         
  .         0  f         h                                   F         H  ~                                 &         (  ^         `                                   >         @  v         x                          "          "  V#         X#  $         $  %         %  &          '  7(         8(  o)         p)  *         *  +         +  -         -  O.         P.  /         /  0         0  1         1  /3         03  g4         h4  5         5  6         6  8         8  G9         H9  :         :  ;         ;  <         <  '>         (>  _?         `?  @         @  A         A  C         C  ?D         @D  wE         xE  F         F  G         G  I          vL  TM          M  xN          N  O          O  P          Q  Q         o   q         q  8r         :r  ps         rs  t         t  u         u  w         w  Px         Rx  y         y  z         z  {         {  0}         2}  h~         j~             ؀         ڀ             H         J                                   (         *  `         b             Ћ         ҋ           
  @         B  x         z                                  "  X                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               +!         ,!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,-C-     C--    -y6-x    y6-E-|    E-ax-\    ax--    --h    -b-    b-Gx-	    Gx--h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     S                       T!  1          1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     :            W  /          t  L                                              <	  /
          u
  h                                 R         S                                   K         L                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                     8  (          l  \	          	  
          
           &  '         '  )         ")  a*         e*  +         +  ,         ,  *.         ..  m/         q/  0         0  1         1  63         :3  y4         }4  5         5  6         7  B8         F8  9         9  :         :  <         <  N=         R=  >         >  ?         ?  A         A  ZB         ^B  C         C  D         D  #F         'F  fG         jG  H         H  I         I  /K         3K  rL         vL  M         M  N         N  =P         ?P  Q         Q  R         R  T         T  IU         KU  V         V  W         W  Y         Y  UZ         WZ  [         [  \         \  ^          ^  a_         c_  `         `  a         a  *c         ,c  md         od  e         e  f         f  6h         8h  yi         {i  j         j  k         l  Bm         Dm  n         n  o         o  q         q  Nr         Pr  s         s  t         t  v         v  Zw         \w  x         x  y         y  #{         %{  f|         h|  }         }  ~         ~  /         1  r         t                        ;         =  ~                    Ç             G         I             ͌         ό             S         U             ّ         ۑ             _         a                        (                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               S!         T!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 [
  {      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -o,-X     o,-%-     %--    -:-x    :-I-|    I-|-\    |--    --h    -Eb-    Eb--	    --h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     P                       B!  1          1  c=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             6         7  m         n                                                                                  6            S  8          v  [            ~                                 a                                     <!         =!  s"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            7         8  o         p                                   O         P             	         	  
         
  /         0  g         h                                   G         H                                   '         (  _         `                                   ?         @  w         x                          "          "  W#         X#  $         $  %         %  &          '  6(         8(  n)         p)  *         *  +         +  -         -  N.         P.  /         /  0         0  1         1  .3         03  f4         h4  5         5  6         6  8         8  F9         H9  ~:         :  ;         ;  <         <  &>         (>  ^?         `?  @         @  A         A  C         C  >D         @D  vE         xE  F         F  G         G  I          *R  S          NS  ,T          rT  PU          U  tV          V  W         o  q         q  9r         :r  qs         rs  t         t  u         u  w         w  Qx         Rx  y         y  z         z  {         {  1}         2}  i~         j~             ـ         ڀ             I         J                                   )         *  a         b             ы         ҋ  	         
  A         B  y         z                        !         "  Y                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               /!         0!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,-[-     [--    -S;-x    S;-I-|    I-;}-\    ;}-9-    9--h    -Wb-    Wb-͂-	    ͂--h
    -%-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     V            !          \!  1          
2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 S	         T	  
         
                                   )         *  `         a              t                                              4            [  7          ~  Z            }              !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            >         C                                   J         O                        
         
  V         [                                 $  b         g                        +         0  n         s                        7         <  z                                  C         H                          "         "  O#         T#  $         $  %         %  '         '  [(         `(  )         )  *         *  $,         ),  g-         l-  .         .  /         /  01         51  s2         x2  3         3  4         4  <6         A6  7         7  8         8  :         
:  H;         M;  <         <  =         =  ?         ?  T@         Y@  A         A  B         B  D         "D  `E         eE  F         F  G         G  )I         .I  lJ         qJ  K          {  |          2}  "~          j~  Z                      ڀ  ʁ         Z             ߖ           "         #  e         f                        .         /  q         r                        :         ;  }         ~                        F         G             ̩         ͩ             R         S             خ         ٮ             ^         _                        '         (  j         k                        3                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               6!         8!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -u,-X     u,-+-     +--    -;-x    ;-kJ-|    kJ-}-\    }-=-    =--P    -ib-    ib-c-	    c--D
    -%-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     Q                       @!  1          1  a=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 5	         6	  l
         m
                        .         /  e         f                                               F  .            g                                              -   !          K!  #"          i"  A#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            9         :  s         t                        !         "  [         \             	         	  	         
  C         D  }         ~                        +         ,  e         f                                   M         N                                   5         6  o         p                        !         !  W"         X"  #         #  $         $  &         &  ?'         @'  y(         z(  )         )  *         *  ',         (,  a-         b-  .         .  /         /  1         1  I2         J2  3         3  4         4  5         5  17         27  k8         l8  9         9  :         :  <         <  S=         T=  >         >  ?         ?  A         A  ;B         <B  uC         vC  D         D  E         E  #G         $G  \H          fT  TU          U  V          V  W          X  X          6Y  $Z         vl  m         m  n         n  "p         $p  \q         ^q  r         r  s         s  
u         u  Dv         Fv  ~w         w  x         x  y         y  ,{         .{  f|         h|  }         }  ~         ~             N         P                      ă             6         8  p         r                                    X         Z             ̎                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               :!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 ]
  }                   
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,-O-     O--    -S;-x    S;-I-|    I-;}-\    ;}-9-    9--h    -]b-    ]b-͂-	    ͂--h
    -%-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     U            !          X!  1          2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 S	         T	  
         
                                   )         *  `         a              u                                              4            [  6          ~  Y            |              !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ?         C                                   K         O                        
         
  W         [                                  $  c         g                        ,         0  o         s                        8         <  {                                 D         H                          "         "  P#         T#  $         $  %         %  '         '  \(         `(  )         )  *         *  %,         ),  h-         l-  .         .  /         /  11         51  t2         x2  3         3  4         4  =6         A6  7         7  8         8  :         
:  I;         M;  <         <  =         =  ?         ?  U@         Y@  A         A  B         B  D         "D  aE         eE  F         F  G         G  *I         .I  mJ         qJ  K                      J  :            r                                 Z             ޖ           !         #  d         f                        -         /  p         r                        9         ;  |         ~                        E         G             ˩         ͩ             Q         S             ׮         ٮ             ]         _                        &         (  i         k                        2                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               8!         <!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -o,-X     o,-%-     %--    -;-x    ;-kJ-|    kJ-}-\    }-=-    =--P    -ib-    ib-c-	    c--D
    -%-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     P                       >!  1          1  ]=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 :	         ;	  q
         r
                        3         4  j         k                                               K  4            m                                              0   !          M!  %"          j"  B#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            8         :  r         t                                  "  Z         \             	         	           
  B         D  |         ~                        *         ,  d         f                                   L         N                                   4         6  n         p                        !         !  V"         X"  #         #  $         $  &         &  >'         @'  x(         z(  )         )  *         *  &,         (,  `-         b-  .         .  /         /  1         1  H2         J2  3         3  4         4  5         5  07         27  j8         l8  9         9  :         :  <         <  R=         T=  >         >  ?         ?   A         A  :B         <B  tC         vC  D         D  E         E  "G         $G  ]H          jZ  X[          [  \          \  ]          ^  ^          :_  (`         vl  m         m  n         n  #p         $p  ]q         ^q  r         r  s         s  u         u  Ev         Fv  w         w  x         x  y         y  -{         .{  g|         h|  }         }  ~         ~             O         P             Ã         ă             7         8  q         r                                    Y         Z             ͎                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               <!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 ]
  }                   
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,--     -#-    #-;-x    ;-I-|    I-|-\    |--    -Y-h    Y-a-    a-͂-	    ͂--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               9!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             6         7  m         n                                  r  N            q                                              !            ?            ]  5          {  S            q         f                        
             A!         B!  x"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            B         C                                   N         O                        
         
  Z         [                        #         $  f         g                        /         0  r         s                        ;         <  ~                                 G         H                          "         "  S#         T#  $         $  %         %  '         '  [(         `(  )         )  *         *  $,         ),  g-         l-  .         .  /         /  01         51  s2         x2  3         3  4         4  <6         A6  7         7  8         8  :         
:  H;         M;  <         <  =         =  ?         ?  T@         Y@  A         A  B         B  D         "D  `E         eE  F         F  G         G  )I          `  a          a  b          c  c          ?d  e          ge  Gf         l  m         m  o         o  Rp         Sp  q         q  r         r  t         t  ^u         _u  v         v  w         w  'y         (y  jz         kz  {         {  |         |  3~         4~  v         w                        ?         @             Ņ         ƅ           	  K         L             ъ         Ҋ             W         X             ݏ         ޏ            !  c           ؔ         ٔ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     E                       .!  s1          1  :=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,-m-     m--    -;-    ;-I-    I-|-d    |--    --p    -9b-    9b-Q-
    Q-_-
    _-ð-t    ð-3׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               1!         >!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             6         7  m         n                                                                                  :            X  >          {  a                                             f                        
             A!         B!  x"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         <  9            |                                  H  E         T#  $         $  %         %  '         '  C(         D(  )         )  *         *  +         +  3-         4-  o.         p.  /         /  0         0  #2         $2  _3         `3  4         4  5         5  7         7  O8         P8  9         9  :         :  <         <  ?=         @=  {>         |>  ?         ?  @         @  /B         0B  kC         lC  D         D  E         E  G          G  [H         \H  I         J  K         K  8M         :M  tN         vN  O         O  P         P  (R         *R  dS         fS  T         T  U         U  W         W  TX         VX  Y         Y  Z         Z  \         
\  D]         F]  ^         ^  _         _  `         `  4b         6b  pc         rc  d         d  e         e  $g         &g  `h         bh  i         i  j         j  l         l  Pm         Rm  n         n  o         o  q         q  @r         Br  |s         ~s  t         t  u         u  0w         2w  lx         nx  y         y  z         z   |         "|  \}         ^}  ~         ~                        L         N             Ą         Ƅ              <         >  x         z                        ,         .  h         j                                   X         Z                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     Q                       R!  1          1  |=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    ~
         
             |      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                      
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     --    -;-x    ;-I-|    I-|-\    |--    -e-h    e-b-    b-͂-	    ͂--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               7!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             6         7  m         n                                  r  M            p                                              !            >            [  3          x  P            m         a                                     <!         =!  s"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            A         C                        
           M         O                        
         
  Y         [                        "         $  e         g                        .         0  q         s                        :         <  }                                 F         H                          "         "  R#         T#  $         $  %         %  '         '  \(         `(  )         )  *         *  %,         ),  h-         l-  .         .  /         /  11         51  t2         x2  3         3  4         4  =6         A6  7         7  8         8  :         
:  I;         M;  <         <  =         =  ?         ?  U@         Y@  A         A  B         B  D         "D  aE         eE  F         F  G         G  *I          f  og          g  h          h  i          j  j          /k  l         l  m         m  o         o  Qp         Sp  q         q  r         r  t         t  ]u         _u  v         v  w         w  &y         (y  iz         kz  {         {  |         |  2~         4~  u         w                        >         @             ą         ƅ           	  J         L             Њ         Ҋ             V         X             ܏         ޏ           !  b           ה         ٔ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     F                       2!  x1          1  A=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,-s-     s--    -;-    ;-I-    I-|-d    |-K-    K--p    -9b-    9b-Q-
    Q-_-
    _-ð-t    ð-3׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               3!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             6         7  m         n                                                                                  6            S  8          v  [            ~                                 a                                     <!         =!  s"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                         !          "  #         T#  $         $  %         %  '         '  B(         D(  ~)         )  *         *  +         +  2-         4-  n.         p.  /         /  0         0  "2         $2  ^3         `3  4         4  5         5  7         7  N8         P8  9         9  :         :  <         <  >=         @=  z>         |>  ?         ?  @         @  .B         0B  jC         lC  D         D  E         E  G          G  ZH         \H  I         J  K         K  9M         :M  uN         vN  O         O  P         P  )R         *R  eS         fS  T         T  U         U  W         W  UX         VX  Y         Y  Z         Z  	\         
\  E]         F]  ^         ^  _         _  `         `  5b         6b  qc         rc  d         d  e         e  %g         &g  ah         bh  i         i  j         j  l         l  Qm         Rm  n         n  o         o  q         q  Ar         Br  }s         ~s  t         t  u         u  1w         2w  mx         nx  y         y  z         z  !|         "|  ]}         ^}  ~         ~                        M         N             ń         Ƅ             =         >  y         z                        -         .  i         j                                   Y         Z                        ٔ  ֕                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     P                       R!  1          1  {=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    ~
         
             |      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                      
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     --    -Y;-x    Y;-I-|    I-A}-\    A}--    --P    -3b-    3b-c-	    c-1-D
    1-9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               6!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 S	         T	  
         
                                   )         *  `         a              t                                              4            [  7          ~  Z            }              !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            >         C                                   J         O                        
         
  V         [                                 $  b         g                        +         0  n         s                        7         <  z                                  C         H                          "         "  O#         T#  $         $  %         %  '         '  [(         `(  )         )  *         *  $,         ),  g-         l-  .         .  /         /  01         51  s2         x2  3         3  4         4  <6         A6  7         7  8         8  :         
:  H;         M;  <         <  =         =  ?         ?  T@         Y@  A         A  B         B  D          @G  -H          pH  ]I          I  J         ng  h         h  i         i  2k         7k  ul         zl  m         m  n          o  >p         Cp  q         q  r         r  t         t  Ju         Ou  v         v  w         w  y         y  Sz         Xz  {         {  |         |  ~         ~  ]         b                        &         +  l         m                        2         3  t         u                        =         >           4  r         w            d  Q                               >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     K                       L!  1          1  m=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 k
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,-s-     s--    -5-    5-D-    D-w-l    w-G4-    G4--x    -ib-	    ib-W-
    W-e-
    e--|    -ֳ-    ֳ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               3!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         #  	          F  ,          i  O            r                                              $	  

          K
  1         r                                   M         N           H  ~                                 #         $  Z                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                        ,            X  A            m                     (#  c$         CF  ~G         G  H         H  I         I  8K         :K  vL         xL  M         M  N         N  0P         2P  nQ         pQ  R         R  S         S  (U         *U  fV         hV  W         W  X         X   Z         "Z  ^[         `[  \         \  ]         ]  _         _  V`         X`  a         a  b         b  d         d  Ne         Pe  f         f  g         g  i         
i  Fj         Hj  k         k  l         l   n         n  >o         @o  |p         ~p  q         q  r         r  6t         8t  tu         vu  v         v  w         w  .y         0y  lz         nz  {         {  |         |  &~         (~  d         f                                    \         ^             ؆         چ             T         V             Ћ         ҋ             L         N             Ȑ         ʐ             C         E                                   9         ;  w         y                        1         3  o         q                        )         +  g         i                        !         #  _         a             ۪         ݪ             W         Y             ӯ         կ             O         Q             ˴                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               4!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    ~
         
             |      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    ~
         
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            0     -,-X     ,-y-     y--    -Y;-x    Y;-I-|    I-A}-\    A}--    --P    -3b-    3b-c-	    c-1-D
    1-9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               5!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 S	         T	  
         
                                   )         *  `         a              u                                              4            [  6          ~  Y            |              !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            ?         C                                   K         O                        
         
  W         [                                  $  c         g                        ,         0  o         s                        8         <  {                                 D         H                          "         "  P#         T#  $         $  %         %  '         '  \(         `(  )         )  *         *  %,         ),  h-         l-  .         .  /         /  11         51  t2         x2  3         3  4         4  =6         A6  7         7  8         8  :         
:  I;         M;  <         <  =         =  ?         ?  U@         Y@  A         A  B         B  D         ng  h         h  i         i  3k         7k  vl         zl  m         m  n          o  ?p         Cp  q         q  r         r  t         t  Ku         Ou  v         v  w         w  y         y  Tz         Xz  {         {  |         |  ~         ~  ^         b                        '         +  k         m                        1         3  s         u                        <         >           4  s         w            Ĕ                        $            T  A            q                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     L                       L!  1          1  n=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 l
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,-y-     y--    -5-    5-D-    D-w-l    w-G4-    G4--x    -ib-	    ib-W-
    W-e-
    e--|    -ֳ-    ֳ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               5!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         #            F  +          i  N            q                                              $	  
          K
  2         r                                   M         N           H  ~                                 #         $  Z                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                 4  	          `	  I
          
  u         (#  d$         CF  G         G  H         H  I         I  9K         :K  wL         xL  M         M  N         N  1P         2P  oQ         pQ  R         R  S         S  )U         *U  gV         hV  W         W  X         X  !Z         "Z  _[         `[  \         \  ]         ]  _         _  W`         X`  a         a  b         b  d         d  Oe         Pe  f         f  g         g  	i         
i  Gj         Hj  k         k  l         l  n         n  ?o         @o  }p         ~p  q         q  r         r  7t         8t  uu         vu  v         v  w         w  /y         0y  mz         nz  {         {  |         |  '~         (~  e         f                                    ]         ^             ن         چ             U         V             ы         ҋ             M         N             ɐ         ʐ             D         E                                   :         ;  x         y                        2         3  p         q                        *         +  h         i                        "         #  `         a             ܪ         ݪ             X         Y             ԯ         կ             P         Q             ̴                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               4!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    ~
         
             |      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    ~
         
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            8     -,-h     ,-y-     y--    -;-    ;-I-    I-|-l    |--    --P    --    -b-h	    b-W-p
    W-e-
    e-ᰳ-    ᰳ-9׳-L                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               3!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             6         7  m         n                                  r  N            q                                              !            ?            ]  5          {  S            q         f                        
             A!         B!  x"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        d#  $         $  %         %  "'         $'  b(         d(  )         )  *         *  ",         $,  b-         d-  .         .  /         /  "1         $1  b2         d2  3         3  4         4  "6         $6  b7         d7  8         8  9         9  ";         $;  b<         d<  =         =  >         >  "@         $@  bA         dA  B         B  C         C  "E         $E  bF         dF  G         G  H         H  "J         $J  cK         dK  L         L  M         M  #O         $O  cP         dP  Q         Q  R         R  #T         $T  cU         dU  V         V  W         W  #Y         $Y  cZ         dZ  [         [  \         \  #^         $^  c_         d_  `         `  a         a  #c         $c  cd         dd  e         e  f         f  #h         $h  ci         di  j         j  k         k  #m         $m  cn         dn  o         ds  t         t  u         u  "w         $w  bx         dx  y         y  z         z  "|         $|  b}         d}  ~          ~                        $  !          d  a                       "         $  b         d                        "         $  b         d                        "         $  b         d                        "         $  b         d                                b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ?!         Z!  1         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    ~
         
             |                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     6
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     --    -;-x    ;-I-|    I-|-\    |-2-    2--h    -{b-    {b-n-	    n--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               4!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             6         7  m         n                                                                                  :            X  >          {  a                                             f                        
             A!         B!  x"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ?         @                                    ?         @                        	          
  ?         @                                    ?         @                                    ?         @                                    ?         @                                    ?         @               !          ,  -          .  .          4/  0          S0  /1          r1  N2         F  H         H  GI         I  J         J  L         L  QM         SM  N         N  O         O  Q         Q  QR         SR  S         S  T         T  V         V  QW         SW  X         X  Y         Y  [         [  Q\         S\  ]         ]  ^         ^  `         `  Qa         Sa  b         b  c         c  e         e  Qf         Sf  g         g  h         h  j         j  Qk         Sk  l         l  m         m  o         o  p         p  #r         $r  cs           א         ؐ             W         X             ו         ؕ             W         X             ך         ؚ             W         X             ן         ؟             W         X             פ         ؤ             W         X             ש         ة             W         X             ׮         خ             W         X                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               7!         J!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     9
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,-y-     y--    -;-    ;-I-    I-|-l    |--    --P    --    -b-h	    b-W-p
    W-e-
    e-۰-    ۰-9׳-L                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               3!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             6         7  m         n                                  r  M            p                                              !            >            [  3          x  P            m         a                                     <!         =!  s"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        d#  $         $  %         %  #'         $'  c(         d(  )         )  *         *  #,         $,  c-         d-  .         .  /         /  #1         $1  c2         d2  3         3  4         4  #6         $6  c7         d7  8         8  9         9  #;         $;  c<         d<  =         =  >         >  #@         $@  cA         dA  B         B  C         C  #E         $E  cF         dF  G         G  H         H  #J         $J  bK         dK  L         L  M         M  "O         $O  bP         dP  Q         Q  R         R  "T         $T  bU         dU  V         V  W         W  "Y         $Y  bZ         dZ  [         [  \         \  "^         $^  b_         d_  `         `  a         a  "c         $c  bd         dd  e         e  f         f  "h         $h  bi         di  j         j  k         k  "m         $m  bn         dn  o         ds  t         t  u         u  #w         $w  cx         dx  y         y  z         z  #|         $|  c}         d}  ~         ~             #         $  c         d                                   $  !          d  a                                 $  c         d                        #         $  c         d                        #         $  c         d                                b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ?!         Z!  1         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    ~
         
             |                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     5
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     --    -;-x    ;-I-|    I-|-\    |-2-    2--h    -b-    b-n-	    n--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               4!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             6         7  m         n                                                                                  6            S  8          v  [            ~                                 a                                     <!         =!  s"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            >         @  ~                                  >         @  ~                      	          
  >         @  ~                                  >         @  ~                                  >         @  ~                                  >         @  ~                                  >         @  ~             !          2  m3          3  4          4  5          5  6          7  7         F  H         H  FI         I  J         J  L         L  RM         SM  N         N  O         O  Q         Q  RR         SR  S         S  T         T  V         V  RW         SW  X         X  Y         Y  [         [  R\         S\  ]         ]  ^         ^  `         `  Ra         Sa  b         b  c         c  e         e  Rf         Sf  g         g  h         h  j         j  Rk         Sk  l         l  m         m  o         o  p         p  "r         $r  bs           ֐         ؐ             V         X             ֕         ؕ             V         X             ֚         ؚ             V         X             ֟         ؟             V         X             ֤         ؤ             V         X             ֩         ة             V         X             ֮         خ             V         X                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               9!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     :
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     -#-    #-6-    6-+E-    +E-yx-l    yx-8-    8--x    -{b-	    {b-]-
    ]-k-
    k--|    -ֳ-    ֳ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               5!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     <            Z  2          x  P                                              A	  3
          z
  l                                !  W         X                                   P         Q                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         <  9            |                                  H  E         zI  J         J  K         K  2M         4M  pN         rN  O         O  P         P  *R         ,R  hS         jS  T         T  U         U  "W         $W  `X         bX  Y         Y  Z         Z  \         \  X]         Z]  ^         ^  _         _  a         a  Pb         Rb  c         c  d         d  
f         f  Hg         Jg  h         h  i         i  k         k  Al         Bl  m         m  n         n  o         o  9q         :q  wr         xr  s         s  t         t  1v         2v  ow         pw  x         x  y         y  ){         *{  g|         h|  }         }  ~         ~  !         "  _         `             ۃ         ܃             W         X             ӈ         Ԉ             O         P             ˍ         ̍  	         
  G         H             Ò         Ē             ?         @  }         ~                        7         8  u         v                        /         0  m         n                        '         (  e         f                                    ]         ^             ٬         ڬ             U         V             ѱ         ұ             L         N             ȶ         ޶             Y                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               6!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    ~
         
             |      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    ~
         
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     r
            0     -,-X     ,--     -#-    #-5-x    5-D-|    D-w-\    w--    --h    -{b-    {b-n-	    n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               6!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         #  	          F  ,          i  O            r                                              $	  

          K
  1         r                                   M         N           H  ~                                 #         $  Z                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            B         C                                   N         O                        
         
  Z         [                        #         $  f         g                        /         0  r         s                        ;         <  ~                                 G         H                          "         "  S#         T#  $         $  %         %  '         '  [(         `(  )         )  *         *  $,         ),  g-         l-  .         .  /         /  01         51  s2         x2  3         3  4         4  <6         A6  7         7  8         8  :         
:  H;         M;  <         <  =         =  ?         ?  T@         Y@  A         A  B         B  D         "D  `E         eE  F          ^  ^          <_  (`          f`  Ra          a  |b          b  c         i  j         j  7l         <l  zm         m  n         n   p         p  Cq         Hq  r         r  s         s  u         u  Ov         Tv  w         w  x         x  z         z  [{         `{  |         |  }         }  $         )  g         l                        0         5  s         x                        <         A                      ǋ           
  H         M             ΐ         Ӑ             T         Y                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               :!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     9
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     -#-    #-6-    6-+E-    +E-yx-l    yx-8-    8--x    -ub-	    ub-]-
    ]-k-
    k--|    -ֳ-    ֳ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               5!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     :            W  /          t  L                                              <	  /
          u
  h                                 R         S                                   K         L                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                         !          "  #          T#  Q$          $  %         zI  J         J  K         K  3M         4M  qN         rN  O         O  P         P  +R         ,R  iS         jS  T         T  U         U  #W         $W  aX         bX  Y         Y  Z         Z  \         \  Y]         Z]  ^         ^  _         _  a         a  Qb         Rb  c         c  d         d  f         f  Ig         Jg  h         h  i         i  k         k  @l         Bl  ~m         m  n         n  o         o  8q         :q  vr         xr  s         s  t         t  0v         2v  nw         pw  x         x  y         y  ({         *{  f|         h|  }         }  ~         ~            "  ^         `             ڃ         ܃             V         X             ҈         Ԉ             N         P             ʍ         ̍           
  F         H                      Ē              >         @  |         ~                        6         8  t         v                        .         0  l         n                        &         (  d         f                                    \         ^             ج         ڬ             T         V             б         ұ             M         N             ɶ         ޶             X                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               4!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    ~
         
             |      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    ~
         
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     q
            0     -,-X     ,--     -#-    #-5-x    5-D-|    D-w-\    w--    --h    -ub-    ub-n-	    n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               6!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         #            F  +          i  N            q                                              $	  
          K
  2         r                                   M         N           H  ~                                 #         $  Z                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            A         C                        
           M         O                        
         
  Y         [                        "         $  e         g                        .         0  q         s                        :         <  }                                 F         H                          "         "  R#         T#  $         $  %         %  '         '  \(         `(  )         )  *         *  %,         ),  h-         l-  .         .  /         /  11         51  t2         x2  3         3  4         4  =6         A6  7         7  8         8  :         
:  I;         M;  <         <  =         =  ?         ?  U@         Y@  A         A  B         B  D         "D  aE         eE  F          c  d          e  e          8f  $g          bg  Nh          h  xi         i  j         j  8l         <l  {m         m  n         n  p         p  Dq         Hq  r         r  s         s  u         u  Pv         Tv  w         w  x         x  z         z  \{         `{  |         |  }         }  %         )  h         l                        1         5  t         x                        =         A             Ë         ǋ           
  I         M             ϐ         Ӑ             U         Y                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               6!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     :
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-;-x    ;-I-|    I-}-\    }-'9-    '9--h    -b-    b-n-	    n--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               :!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             6         7  m         n                                  r  N            q                                              !            ?            ]  5          {  S            q         f                        
             A!         B!  x"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            :         <  v         x                        *         ,  f         h             	         	             V         X                        
           F         H                                   6         8  r         t                        &         (  b         d                                     R!         T!  "         "  #         #  %         %  C&         D&  '         '  (         (  )         )  3+         4+  o,         p,  -         -  .         .  #0         $0  _1         `1  2         2  3         3  5         5  O6         P6  7         7  8         8  :         :  ?;         @;  {<         |<  =         =  >         >  /@         0@  kA         lA  B         B  C         C  E          E  [F         \F  G          G  H          I  I          +J  K          KK  %L          kL  EM         m  o         o  Op           ֕         ؕ             N         P             ƚ         Ț             >         @  z         |                        .         0  j         l                                    Z         \             Ҫ         Ԫ             J         L             ¯         į              :         <  v         x                        *         ,  f         h                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               =!         R!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     +
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-;-x    ;-}J-|    }J-}-\    }-]6-    ]6--h    -ub-    ub-wx-	    wx--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               >!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 5	         6	  l
         m
                        .         /  e         f                                               F  .            g                                              -   !          K!  #"          i"  A#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            B         C                                   N         O                        
         
  Z         [                        #         $  f         g                        /         0  r         s                        ;         <  ~                                 G         H                          "         "  S#         T#  $         $  %         F  G         G  -I         .I  pJ         qJ  K         K  L         L  9N         :N  |O         }O  P         P  R         R  ES         FS  T         T  U         U  W         W  QX         RX  Y         Y  Z         Z  \         \  ]]         ^]  ^         ^  _         _  &a         'a  ib         jb  c         c  d         d  2f         3f  ug         vg  h         h  i         i  >k         ?k  l         l  m         m  o         o  Fp         Kp  q         q  r         r  t         t  Ru         Wu  v         v  w         w  y          y  ^z         cz  {         {  |         |  '~         ,~  j         o                        3         8  v         {                        ?         D             Ŋ         ʊ             K          P  2            b                                             ұ             S         X             ٶ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               3!         B!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 m
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-;-x    ;-I-|    I-|-\    |-'9-    '9--h    -b-    b-n-	    n--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               <!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             6         7  m         n                                  r  M            p                                              !            >            [  3          x  P            m         a                                     <!         =!  s"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ;         <  w         x                        +         ,  g         h             	         	             W         X                                   G         H                                   7         8  s         t                        '         (  c         d                                     S!         T!  "         "  #         #  %         %  B&         D&  ~'         '  (         (  )         )  2+         4+  n,         p,  -         -  .         .  "0         $0  ^1         `1  2         2  3         3  5         5  N6         P6  7         7  8         8  :         :  >;         @;  z<         |<  =         =  >         >  .@         0@  jA         lA  B         B  C         C  E          E  ZF         \F  G          M  eN          N  O          O  P          P  Q          R  R         m  o         o  Np           ו         ؕ             O         P             ǚ         Ț             ?         @  {         |                        /         0  k         l                                    [         \             Ӫ         Ԫ             K         L             ï         į              ;         <  w         x                        +         ,  g         h                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               =!         R!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     *
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-;-x    ;-}J-|    }J-}-\    }-c6-    c6--h    -{b-    {b-}x-	    }x--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               :!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 :	         ;	  q
         r
                        3         4  j         k                                               K  4            m                                              0   !          M!  %"          j"  B#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            A         C                        
           M         O                        
         
  Y         [                        "         $  e         g                        .         0  q         s                        :         <  }                                 F         H                          "         "  R#         T#  $         $  %         F  G         G  ,I         .I  oJ         qJ  K         K  L         L  8N         :N  {O         }O  P         P  R         R  DS         FS  T         T  U         U  W         W  PX         RX  Y         Y  Z         Z  \         \  \]         ^]  ^         ^  _         _  %a         'a  hb         jb  c         c  d         d  1f         3f  tg         vg  h         h  i         i  =k         ?k  l         l  m         m  o         o  Gp         Kp  q         q  r         r  t         t  Su         Wu  v         v  w         w  y          y  _z         cz  {         {  |         |  (~         ,~  k         o                        4         8  w         {                        @         D             Ɗ         ʊ  	           L          @  "          p  R                      Ж                        ұ             T         X             ڶ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               5!         F!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 m
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-6-x    6-7E-|    7E-x-\    x-s#-    s#--h    -b-    b-n-	    n--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               <!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     <            Z  2          x  P                                              A	  8
          z
  q                                !  W         X                                   P         Q                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            8         :  r         t                                  "  Z         \             	         	           
  B         D  |         ~                        *         ,  d         f                                   L         N                                   4         6  n         p                        !         !  V"         X"  #         #  $         H  J         J  WK         XK  L         L  M         M  O         O  ?P         @P  yQ         zQ  R         R  S         S  'U         (U  aV         bV  W         W  X         X  Z         Z  I[         J[  \         \  ]         ]  ^         ^  1`         2`  ka         la  b         b  c         c  e         e  Sf         Tf  g         g  h         h  j         j  ;k         <k  ul         vl  m         No  p         p  q         q  r         r  5t         6t  ou         pu  v         v  w         w  y         y  Wz         Xz  {         {  |         |  ~         ~  ?         @  y         z                        '         (  a         b             Շ         և             I         J                                   1         2  k         l             ߒ                     p  P            v                        ¢                               >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               @!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     +
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-6-x    6-D-|    D-w-\    w--    --h    -cb-    cb-}x-	    }x--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               8!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         #            F  '          i  J            m                                              $	  
          K
  ,         r                                   M         N           H  ~                                 #         $  Z                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                     R  B            x                                 %  6'         7'  y(         z(  )         )  *          +  B,         C,  -         -  .         .  0         0  N1         O1  2         2  3         3  5         5  Z6         [6  7         7  8         8  #:         $:  f;         g;  <         <  =         =  /?         0?  r@         s@  A         A  B         B  ;D         <D  ~E         E  F         F  H         H  GI         HI  J         J  K         K  M         M  ON         TN  O         O  P         P  R         R  [S         `S  T         T  U         U  $W         )W  gX         lX  Y         Y  Z         Z  0\         5\  s]         x]  ^         ^  _         _  <a         Aa  b         b  c         c  e         
e  Hf         Mf  g         g  h         h  j         j  Tk         Yk  l         l  m         m  o         "o  `p         ep  q         q  r         r  )t         .t  lu         qu  v         v  w         w  5y         :y  xz         }z  {         {  |         }  A~         F~             ǀ         ̀  
           M         R             Ӆ         ؅             Y         ^             ߊ           "         '  e         j                        .         3  q         v                        :         ?  }                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ,!         4!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 m
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -;-    ;-6-x    6-1E-|    1E-x-\    x-1)-    1)--h    -b-    b-n-	    n--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               >!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     :            W  /          t  L                                              <	  2
          u
  k                                 R         S                                   K         L                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            9         :  s         t                        !         "  [         \             	         	  	         
  C         D  }         ~                        +         ,  e         f                                   M         N                                   5         6  o         p                        !         !  W"         X"  #         #  $         H  J         J  VK         XK  L         L  M         M  O         O  >P         @P  xQ         zQ  R         R  S         S  &U         (U  `V         bV  W         W  X         X  Z         Z  H[         J[  \         \  ]         ]  ^         ^  0`         2`  ja         la  b         b  c         c  e         e  Rf         Tf  g         g  h         h   j         j  :k         <k  tl         vl  m         No  p         p  q         q  r         r  4t         6t  nu         pu  v         v  w         w  y         y  Vz         Xz  {         {  |         |  ~         ~  >         @  x         z                        &         (  `         b             ԇ         և             H         J                                   0         2  j         l             ޒ                     .            T  4          z  Z                      ƨ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               >!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     *
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-6-x    6-D-|    D-w-\    w--    --h    -ib-    ib-}x-	    }x--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               4!         @!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         #            F  &          i  I            l                                              $	  
          K
  -         r                                   M         N           H  ~                                 #         $  Z                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         *            `  P                                             %  5'         7'  x(         z(  )         )  *          +  A,         C,  -         -  .         .  
0         0  M1         O1  2         2  3         3  5         5  Y6         [6  7         7  8         8  ":         $:  e;         g;  <         <  =         =  .?         0?  q@         s@  A         A  B         B  :D         <D  }E         E  F         F  H         H  FI         HI  J         J  K         K  M         M  PN         TN  O         O  P         P  R         R  \S         `S  T         T  U         U  %W         )W  hX         lX  Y         Y  Z         Z  1\         5\  t]         x]  ^         ^  _         _  =a         Aa  b         b  c         c  e         
e  If         Mf  g         g  h         h  j         j  Uk         Yk  l         l  m         m  o         "o  ap         ep  q         q  r         r  *t         .t  mu         qu  v         v  w         w  6y         :y  yz         }z  {         {  |         }  B~         F~             Ȁ         ̀             N         R             ԅ         ؅             Z         ^                        #         '  f         j                        /         3  r         v                        ;         ?  ~                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               .!         8!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 m
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     ---X     ---     -_-    _-;-x    ;-GJ-|    GJ-}-\    }--    --h    -{b-    {b-kx-	    kx--h
    -Q׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               S!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 S	         T	  
         
                                   )         *  `         a              p                                  "            P  (          ~  V                                      !          &"  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            7         8  o         p                                   O         P             	         	  
         
  /         0  g         h                                   G         H                                   '         (  _         `                                   ?         @  w         x                          "          "  W#         X#  $         $  %         %  &          '  6(         8(  n)         p)  *         *  +         +  -         -  N.         P.  /         /  0         0  1         1  .3         03  f4         h4  5         5  6         6  8         8  F9         H9  ~:         :  ;         ;  <         <  &>         (>  ^?         `?  @         @  A         A  C         C  >D         @D  vE         xE  F         F  G         G  I         I  K         K  ML         NL  M         M  N         N  O         O  -Q         .Q  eR         fR  S         S  T         T  V         V  EW         FW  }X         ~X  Y         Y  Z         Z  %\         &\  ]]         ^]  ^         ^  _         _  a         a  =b         >b  uc         vc  d         d  e         e  g         g  Uh         Vh  i         i  j         j  k         k  5m         6m  mn          ^  K                        ы          '            j  W                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               4!         :!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 [
  {      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-<-x    <-J-|    J-}-\    }-8-    8--h    -{b-    {b-͂-	    ͂--h
    -Q׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               6!         8!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 D	         E	  {
         |
                        =         >  t         u                                               U  <            u                                               <   !          Z!  2"          x"  P#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            B         C                                   N         O                        
         
  Z         [                        #         $  f         g                        /         0  r         s                        ;         <  ~                                 G         H                          "         "  S#         T#  $         $  %         %  '         '  _(         `(  )         )  *         *  (,         ),  k-         l-  .         .  /         /  41         51  w2         x2  3         3  4         4  @6         A6  7         7  8         8  	:         
:  L;         M;  <         <  =         =  ?         ?  X@         Y@  A         A  B         B  !D         "D  dE         eE  F         F  G         G  -I         .I  pJ         qJ  K          b  c          c  d          d  e          f  f          Dg  $h         :  x         }                        A         F             ǚ         ̚  
           M         R             ӟ         ؟             Y         ^             ߤ           "         '  e         j                        .         3  q         v                        :         ?  }                    Ų             F         K             ̷         ѷ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               2!         @!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -S-    S-;-x    ;-GJ-|    GJ-}-\    }--    --h    -b-    b-kx-	    kx--h
    -Q׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               O!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 S	         T	  
         
                                   )         *  `         a              p                                              L  $          y  Q            |                          !          %"  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            6         8  n         p                                   N         P             	         	  
         
  .         0  f         h                                   F         H  ~                                 &         (  ^         `                                   >         @  v         x                          "          "  V#         X#  $         $  %         %  &          '  7(         8(  o)         p)  *         *  +         +  -         -  O.         P.  /         /  0         0  1         1  /3         03  g4         h4  5         5  6         6  8         8  G9         H9  :         :  ;         ;  <         <  '>         (>  _?         `?  @         @  A         A  C         C  ?D         @D  wE         xE  F         F  G         G  I         I  K         K  LL         NL  M         M  N         N  O         O  ,Q         .Q  dR         fR  S         S  T         T  V         V  DW         FW  |X         ~X  Y         Y  Z         Z  $\         &\  \]         ^]  ^         ^  _         _  a         a  <b         >b  tc         vc  d         d  e         e  g         g  Th         Vh  i         i  j         j  k         k  4m         6m  ln                        ݐ          3             v  c          ?  ,                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               6!         >!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 [
  {      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-;-x    ;-J-|    J-}-\    }-8-    8--h    -ub-    ub-͂-	    ͂--h
    -Q׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               5!         6!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                               	  ?	         @	  v
         w
                        8         9  o         p                                               P  6            o                                              5   !          R!  *"          o"  G#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            A         C                        
           M         O                        
         
  Y         [                        "         $  e         g                        .         0  q         s                        :         <  }                                 F         H                          "         "  R#         T#  $         $  %         %  '         '  ^(         `(  )         )  *         *  ',         ),  j-         l-  .         .  /         /  31         51  v2         x2  3         3  4         4  ?6         A6  7         7  8         8  :         
:  K;         M;  <         <  =         =  ?         ?  W@         Y@  A         A  B         B   D         "D  cE         eE  F         F  G         G  ,I         .I  oJ         qJ  K          jh  Ji          i  pj          j  k          k  l          m  m         :  y         }                        B         F             Ț         ̚             N         R             ԟ         ؟             Z         ^                        #         '  f         j                        /         3  r         v                        ;         ?  ~                    Ų             G         K             ͷ         ѷ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               0!         <!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -;-    ;-#;-x    #;-I-|    I-}-\    }--    --h    -]b-    ]b--	    --h
    -K׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               E!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             6         7  m         n                                  r  V            y                                              !            ?            ]  5          {  S            q         f                        
             A!         B!  x"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            B         C                                   N         O                        
         
  Z         [                        #         $  f         g                        /         0  r         s                        ;         <  ~                                 G         H                          "         "  S#         T#  $         $  %         %  '         '  [(         `(  )         )  *         *  $,         ),  g-         l-  .         .  /         /  01         51  s2         x2  3         3  4         4  <6         A6  7         7  8         8  :         
:  H;         M;  <         <  =         =  ?         ?  T@         Y@  A         A  B         B  D         "D  `E         eE  F         F  G         G  )I         zk  l         l  m          n  Bo         Co  p         p  q         q  s         s  Nt         Ot  u         u  v         v  x         x  Zy         [y  z         z  {         {  #}         $}  f~         g~                        /         0  r         s                        ;         <  ~                                 G         H             ͎         Ύ             S         T             ٓ          ړ                        "            F  *          j  N                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               +!         ,!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -;-    ;-#;-    #;-I-    I-}-d    }--    --p    -b-    b-i-
    i-w-
    w-ᰳ-t    ᰳ-K׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               E!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             6         7  m         n                                                                                  :            X  6          {  Y            |                                 f                        
             A!         B!  x"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                        C  @                                    	         T#  $         $  %         %  &         &  3(         4(  k)         l)  *         *  +         +  -         -  K.         L.  /         /  0         0  1         1  +3         ,3  c4         d4  5         5  6         6  8         8  C9         D9  {:         |:  ;         ;  <         <  #>         $>  [?         \?  @         @  A         A  C         C  ;D         <D  sE         tE  F         F  G         G  I         I  RJ         TJ  K         K  L         L  M         M  2O         4O  jP         lP  Q         Q  R         R  T         T  JU         LU  V         V  W         W  X         X  *Z         ,Z  b[         d[  \         \  ]         ]  
_         _  B`         D`  za         |a  b         b  c         c  "e         $e  Zf         \f  g         g  h         h  j         j  :k         <k  rl         tl  m         m  n         n  p         p  Rq         Tq  r         r  s         s  t         t  2v         4v  jw         lw  x         x  y         y  {         {  J|         L|  }         }  ~         ~             *         ,  b         d             ҄         Ԅ  
           B         D  z         |                        "         $  Z         \             ʏ         ̏                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               9!         J!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    ~
         
             |      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     &
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-;-x    ;-I-|    I-}-\    }--    --h    -cb-    cb--	    --h
    -K׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               A!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             6         7  m         n                                  r  U            x                                              !            >            [  3          x  P            m         a                                     <!         =!  s"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            A         C                        
           M         O                        
         
  Y         [                        "         $  e         g                        .         0  q         s                        :         <  }                                 F         H                          "         "  R#         T#  $         $  %         %  '         '  \(         `(  )         )  *         *  %,         ),  h-         l-  .         .  /         /  11         51  t2         x2  3         3  4         4  =6         A6  7         7  8         8  :         
:  I;         M;  <         <  =         =  ?         ?  U@         Y@  A         A  B         B  D         "D  aE         eE  F         F  G         G  *I         zk  l         l  m          n  Ao         Co  p         p  q         q  
s         s  Mt         Ot  u         u  v         v  x         x  Yy         [y  z         z  {         {  "}         $}  e~         g~                        .         0  q         s                        :         <  }                                 F         H             ̎         Ύ             R         T             ؓ            r                      ֛              ޝ                               >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               -!         0!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -G-    G-;-    ;-I-    I-}-d    }--    --p    -b-    b-i-
    i-w-
    w-۰-t    ۰-K׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               I!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             6         7  m         n                                                                                  6            S  0          v  S            v                                 a                                     <!         =!  s"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         O  L                        	          
            [  X         T#  $         $  %         %  &         &  2(         4(  j)         l)  *         *  +         +  -         -  J.         L.  /         /  0         0  1         1  *3         ,3  b4         d4  5         5  6         6  
8         8  B9         D9  z:         |:  ;         ;  <         <  ">         $>  Z?         \?  @         @  A         A  C         C  :D         <D  rE         tE  F         F  G         G  I         I  SJ         TJ  K         K  L         L  M         M  3O         4O  kP         lP  Q         Q  R         R  T         T  KU         LU  V         V  W         W  X         X  +Z         ,Z  c[         d[  \         \  ]         ]  _         _  C`         D`  {a         |a  b         b  c         c  #e         $e  [f         \f  g         g  h         h  j         j  ;k         <k  sl         tl  m         m  n         n  p         p  Sq         Tq  r         r  s         s  t         t  3v         4v  kw         lw  x         x  y         y  {         {  K|         L|  }         }  ~         ~             +         ,  c         d             ӄ         Ԅ             C         D  {         |                        #         $  [         \             ˏ         ̏                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               9!         J!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    ~
         
             |      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     %
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-q;-x    q;-J-|    J-Y}-\    Y}-/-    /--P    -{b-    {b-c-	    c-C-D
    C-E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               <!         @!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 S	         T	  
         
                                   )         *  `         a              |                                              4            [  ?          ~  b                          !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            >         C                                   J         O                        
         
  V         [                                 $  b         g                        +         0  n         s                        7         <  z                                  C         H                          "         2  3         3  	5         5  L6         Q6  7         7  8         8  :         :  X;         ];  <         <  =         =  !?         &?  d@         i@  A         A  B         B  -D         2D  pE         uE  F         F  G         G  9I         >I  |J         J  K         K  M         M  EN         JN  O         O  P         P  R         R  QS         VS  T         T  U         U  W         l  n         n  Uo         Zo  p         p  q         q   s         !s  bt         ct  u         u  v         v  *x         +x  my         ny  z         z  {         {  6}         7}  y~         z~                         B         C             Ȅ         Ʉ             N         O             ԉ         Չ             Z         [                        #          Ε              ϗ                      4            V  5         v                        >                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               4!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 i
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    <     -,-p     ,--     -;-    ;-6-    6-D-    D-w-t    w-U2-     U2-9-X    9--    -b-p	    b-c-x
    c-q-
    q--    -ֳ-\    ֳ--<                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               C!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         #            F  $          i  G            j                                              $	  
          K
  )         r                                   M         N           H  ~                                 #         $  Z                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        "  #         #  !%         #%  [&         ]&  '         '  (         (  	*         *  C+         E+  },         ,  -         -  .         .  +0         -0  e1         g1  2         2  3         3  5         5  M6         O6  7         7  8         8  9         9  5;         7;  o<         q<  =         =  >         >  @         @  WA         YA  B         B  C         C  E         E  ?F         h  8i         :i  rj         tj  k         k  l         l   n         "n  Zo         \o  p         p  q         q  s         
s  Bt         Dt  |u         ~u  v         v  w         w  *y         ,y  dz         fz  {         {  |         |  ~         ~  L         N                                   4         6  n         p                                   V         X             ʌ         ̌             =         ?  w         y                        #         %  ]         _             ї         ӗ             E          G  >            x                                  /  &         i             ۢ         ݢ             O         Q             ç         ŧ             7         9  q         s                                 !  Y         [             Ͳ                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               B!         L!  1         1  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    ~
         
             |                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    ~
         
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     a
            0     -,-X     ,--     -#-    #-q;-x    q;-J-|    J-Y}-\    Y}-/-    /--P    -{b-    {b-c-	    c-C-D
    C-E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               9!         >!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 S	         T	  
         
                                   )         *  `         a              }                                              4            [  >          ~  a                          !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            ?         C                                   K         O                        
         
  W         [                                  $  c         g                        ,         0  o         s                        8         <  {                                 D         H                          "         2  3         3  
5         5  M6         Q6  7         7  8         8  :         :  Y;         ];  <         <  =         =  "?         &?  e@         i@  A         A  B         B  .D         2D  qE         uE  F         F  G         G  :I         >I  }J         J  K         K  M         M  FN         JN  O         O  P         P  R         R  RS         VS  T         T  U         U  W         l  n         n  Vo         Zo  p         p  q         q  s         !s  at         ct  u         u  v         v  )x         +x  ly         ny  z         z  {         {  5}         7}  x~         z~                         A         C             Ǆ         Ʉ  
           M         O             Ӊ         Չ             Y         [             ߎ           "          x  W            y                      ޞ               ߠ         v                        =                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               4!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 i
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    <     -,-p     ,--     -A-    A-6-    6-D-    D-w-t    w-[2-     [2-9-X    9--    -b-p	    b-c-x
    c-q-
    q--    -ֳ-\    ֳ--<                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               G!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         #             F  #          i  F            i                                              $	  
          K
  *         r                                   M         N           H  ~                                 #         $  Z                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        "  #         #  "%         #%  \&         ]&  '         '  (         (  
*         *  D+         E+  ~,         ,  -         -  .         .  ,0         -0  f1         g1  2         2  3         3  5         5  N6         O6  7         7  8         8  9         9  6;         7;  p<         q<  =         =  >         >  @         @  XA         YA  B         B  C         C  E         E  @F         h  9i         :i  sj         tj  k         k  l         l  !n         "n  [o         \o  p         p  q         q  	s         
s  Ct         Dt  }u         ~u  v         v  w         w  +y         ,y  ez         fz  {         {  |         |  ~         ~  M         N                                   5         6  o         p                                   W         X             ˌ         ̌             >         ?  x         y                        $         %  ^         _             җ         ӗ             F         G                                   .         /  h          i  `                      ݢ  ԣ                      Q  H           ħ         ŧ             8         9  r         s                                  !  Z         [             β                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               D!         L!  1         1  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    ~
         
             |                              b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    ~
         
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     b
            4     -,-`     ,--     -#-    #-;-    ;-I-    I-}-d    }--    --p    -b-    b-c-
    c-q-
    q--t    -E׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ;!         >!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             6         7  m         n                                  r  V            y                                              !            ?            ]  5          {  S            q         f                        
             A!         B!  x"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       	          L  ;          ~  m                     %  F&         K&  '         '  (         (  *         *  R+         W+  ,         ,  -         -  /          /  ^0         c0  1         1  2         2  '4         ,4  j5         o5  6         6  7         7  39         89  v:         {:  ;         ;  <         =  ?>         D>  ?         ?  @         @  B         B  KC         PC  D         D  E         E  G         G  WH         \H  I         I  J         J   L         %L  cM         hM  N         N  O         O  ,Q         1Q  oR         tR  S         S  T         T  8V         =V  {W         W  X         X  Z         Z  D[         I[  \         \  ]         ]  _         _  P`         U`  a         a  b         b  d         d  \e         ae  f         f  g         g  %i         *i  hj         mj  k         k  l         l  5n         6n  xo         yo  p         p  q         q  As         Bs  t         t  u         u  
w         w  Mx         Nx  y         y  z         z  |         |  Y}         Z}  ~         ~             "         #  e         f                        .         /  q         r                        :         ;  }         ~                        F         G             ̒         ͒             N         S                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ?!         R!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    ~
         
             |      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     >
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-;-x    ;-I-|    I-}-\    }--    --h    -b-    b-!n-	    !n--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               >!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             6         7  m         n                                                                                  :            X  6          {  Y            |                                 f                        
             A!         B!  x"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ;         <  w         x                        +         ,  g         h             	         	             W         X                        
           F         H                                   6         8  r         t                        &         (  b         d                                     R!         T!  "         "  #         #  %          <  m=          =  >          >  ?          @  @          .A  B         :H  uI         vI  J         J  K         K  )M         *M  eN         fN  O         O  P         P  R         R  US         VS  T         T  U         U  	W         
W  EX         FX  Y         Y  Z         Z  [         [  5]         6]  q^         r^  _         _  `         `  %b         &b  ac         bc  d         d  e         e  g         g  Qh         Rh  i         i  j         j  l         l  Am         Bm  |n         ~n  o         o  p         p  0r         2r  ls         ns  t         t  u         u   w         "w  \x         ^x  y         y  z         z  |         |  L}         N}  ~         ~                         <         >  x         z                        ,         .  h         j                                   X         Z             Џ         ҏ             H                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ?!         R!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     9
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -)-    )-;-    ;-I-    I-|-d    |--    --p    -b-    b-c-
    c-q-
    q--t    -E׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               =!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             6         7  m         n                                  r  U            x                                              !            >            [  3          x  P            m         a                                     <!         =!  s"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                  F   5!          x!  g"          "  #         %  G&         K&  '         '  (         (  *         *  S+         W+  ,         ,  -         -  /          /  _0         c0  1         1  2         2  (4         ,4  k5         o5  6         6  7         7  49         89  w:         {:  ;         ;  <         =  @>         D>  ?         ?  @         @  	B         B  LC         PC  D         D  E         E  G         G  XH         \H  I         I  J         J  !L         %L  dM         hM  N         N  O         O  -Q         1Q  pR         tR  S         S  T         T  9V         =V  |W         W  X         X  Z         Z  E[         I[  \         \  ]         ]  _         _  Q`         U`  a         a  b         b  d         d  ]e         ae  f         f  g         g  &i         *i  ij         mj  k         k  l         l  4n         6n  wo         yo  p         p  q         q  @s         Bs  t         t  u         u  	w         w  Lx         Nx  y         y  z         z  |         |  X}         Z}  ~         ~             !         #  d         f                        -         /  p         r                        9         ;  |         ~                        E         G             ˒         ͒             O         S                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               =!         R!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    ~
         
             |      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     =
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-;-x    ;-I-|    I-|-\    |--    --h    -b-    b-!n-	    !n--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               @!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             6         7  m         n                                                                                  6            S  0          v  S            v                                 a                                     <!         =!  s"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            :         <  v         x                        *         ,  f         h             	         	             V         X                                   G         H                                   7         8  s         t                        '         (  c         d                                     S!         T!  "         "  #         #  %          XB  ?C          C  iD          D  E          E  F           G  G         :H  tI         vI  J         J  K         K  (M         *M  dN         fN  O         O  P         P  R         R  TS         VS  T         T  U         U  W         
W  DX         FX  Y         Y  Z         Z  [         [  4]         6]  p^         r^  _         _  `         `  $b         &b  `c         bc  d         d  e         e  g         g  Ph         Rh  i         i  j         j  l         l  @m         Bm  }n         ~n  o         o  p         p  1r         2r  ms         ns  t         t  u         u  !w         "w  ]x         ^x  y         y  z         z  |         |  M}         N}  ~         ~                        =         >  y         z                        -         .  i         j                                   Y         Z             я         ҏ             I                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               C!         V!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     :
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     -#-    #-6-    6-+E-    +E-yx-l    yx-u9-    u9--x    -ub-	    ub-]-
    ]-k-
    k--|    -ֳ-    ֳ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               9!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     <            Z  2          x  P                                              A	  +
          z
  d                                !  W         X                                   P         Q                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                        >  6          |  t                                 D%  &         I  J         J  4L         5L  wM         xM  N         N  O         O  @Q         AQ  R         R  S         S  	U         
U  LV         MV  W         W  X         X  Z         Z  X[         Y[  \         \  ]         ]  !_         "_  d`         e`  a         a  b         b  -d         .d  pe         qe  f         f  g         g  9i         :i  |j         }j  k         k  m         m  An         Fn  o         o  p         p  r         r  Qs         Rs  t         t  u         u  w         w  ]x         ^x  y         y  z         z  &|         '|  i}         j}  ~         ~             2         3  u         v                        >         ?             Ĉ         ň             J         K             Ѝ         э             V         W             ܒ         ݒ              ^         c                        +         ,  j         o                        3         8  v         {                        ?         D             ť         ʥ             K         P             Ѫ         ֪             W         \             ݯ                     %  c         h                        ,         1  o         t                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               8!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    ~
         
             |      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    ~
         
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            0     -,-X     ,--     -)-    )-5-x    5-D-|    D-w-\    w-w-    w--h    -{b-    {b-n-	    n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               :!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         #            F  $          i  G            j                                              $	  
          K
  )         r                                   M         N           H  ~                                 #         $  Z                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            =         >  {         |                        5         6  s         t             	         	  -         .  k         l                        %         &  c         d                                   [         \                                   S         T                                   K          L   !         !  "         "  $         $  C%          %  h&          &  '          '  (          
)  )          8*   +         G  H         H  0J         2J  nK         pK  L         L  M         M  (O         *O  fP         hP  Q         Q  R         R   T         "T  ^U         `U  V         V  W         W  Y         Y  VZ         XZ  [         [  \         \  ^         ^  N_         P_  `         `  a         a  c         
c  Fd         Hd  e         e  f         f   h         h  >i         @i  |j         ~j  k         k  l         n  o         o  *q         ,q  hr         jr  s         s  t         t  "v         $v  `w         bw  x         x  y         y  {         {  X|         Z|  }         }  ~         ~             P         R             ̃         ΃  
           H         J             Ĉ         ƈ             @         B  ~                                 8         :  v         x                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               6!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     9
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     -)-    )-6-    6-+E-    +E-yx-l    yx-{9-    {9--x    -ub-	    ub-]-
    ]-k-
    k--|    -ֳ-    ֳ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ;!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     :            W  /          t  L            |                                  <	  '
          u
  `                                 R         S                                   K         L                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         6  .          t  l            	          	  
          .  &         D%  &         I  J         J  3L         5L  vM         xM  N         N  O         O  ?Q         AQ  R         R  S         S  U         
U  KV         MV  W         W  X         X  Z         Z  W[         Y[  \         \  ]         ]   _         "_  c`         e`  a         a  b         b  ,d         .d  oe         qe  f         f  g         g  8i         :i  {j         }j  k         k  m         m  Bn         Fn  o         o  p         p  r         r  Ps         Rs  t         t  u         u  w         w  \x         ^x  y         y  z         z  %|         '|  h}         j}  ~         ~             1         3  t         v                        =         ?             È         ň             I         K             ύ         э             U         W             ے         ݒ              _         c                        *         ,  k         o                        4         8  w         {                        @         D             ƥ         ʥ  	           L         P             Ҫ         ֪             X         \             ޯ           !         %  d         h                        -         1  p         t                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               8!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    ~
         
             |      
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    ~
         
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            0     -,-X     ,--     -)-    )-5-x    5-D-|    D-w-\    w-w-    w--h    -{b-    {b-n-	    n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               <!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         #             F  #          i  F            i                                              $	  
          K
  *         r                                   M         N           H  ~                                 #         $  Z                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            <         >  z         |                        4         6  r         t             	         	  ,         .  j         l                        $         &  b         d                                   Z         \                                   R         T                                   J          L   !         !  "         "  $         $  B%          f+  N,          ,  |-          -  .          .  /          0  1         G  H         H  1J         2J  oK         pK  L         L  M         M  )O         *O  gP         hP  Q         Q  R         R  !T         "T  _U         `U  V         V  W         W  Y         Y  WZ         XZ  [         [  \         \  ^         ^  O_         P_  `         `  a         a  	c         
c  Gd         Hd  e         e  f         f  h         h  ?i         @i  }j         ~j  k         k  l         n  o         o  +q         ,q  ir         jr  s         s  t         t  #v         $v  aw         bw  x         x  y         y  {         {  Y|         Z|  }         }  ~         ~             Q         R             ̓         ΃             I         J             ň         ƈ             A         B                                   9         :  w         x                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               8!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     :
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     --    -;-x    ;-I-|    I-|-\    |-7-    7--h    -{b-    {b-	n-	    	n--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               4!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             6         7  m         n                                  r  V            y                                              !            ?            ]  5          {  S            q         f                        
             A!         B!  x"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            >         @  ~                                  >         @  ~                      	          
  >         @  ~                                  >         @  ~                                  >         @  ~                                  >         @  ~                                  >         @  ~             !         !  "          #  >$         @$  ~%          h1  P2          2  ~3          3  4          4  5           6  7         K  L         L  /N         0N  oO         pO  P         P  Q         Q  /S         0S  oT         pT  U         U  V         V  /X         0X  oY         pY  Z         Z  [         [  /]         0]  o^         p^  _         _  `         `  /b         0b  oc         pc  d         d  e         e  /g         0g  oh         ph  i         i  j         j  /l         0l  om         pm  n         n  o         o  /q         0q  or           2         4  r         t                        2         4  r         t                        2         4  r         t                        2         4  r         t                        2         4  r         t                        2         4  r         t                        2         4  r         t                        2                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               7!         J!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     +
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-<-x    <-J-|    J-}-\    }--    --h    -{b-    {b-qx-	    qx--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               8!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      &         '  ]	         ^	  
         
                         V         W                                              5  %          n  ^                                              7             U   -!          s!  K"          "  i#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ?         @                                    ?         @                        	          
  ?         @                                    ?         @                                    ?         @                                    ?         @                                    ?         @               !         !  "          #  ?$         @$  %         %  &         &  '          (  ?)         @)  *         *  +         +  ,          -  ?.         @.  /         /  0         0  1          2  ?3         @3  4         4  5         5  6          7  ?8         @8  9         9  :         :  ;          <  ?=         @=  >         >  ?         ?  @          A  ?B         @B  C         C  D         D  E          F  ?G         @G  H         H  I         I  J          K  >L         @L  ~M         M  N         N  O          P  >Q         @Q  ~R         R  S         S  T          U  >V         @V  ~W         W  X         X  Y          Z  >[         @[  ~\         \  ]         ]  ^          _  >`         @`  ~a         a  b         b  c          d  >e         @e  ~f         f  g         g  h          i  >j         @j  ~k         k  l         l  m          {  |          Z|  R}          }  ~          ~                          ^         `                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ;!         J!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 m
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     --    -;-x    ;-I-|    I-|-\    |-7-    7--h    -b-    b-	n-	    	n--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               4!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             6         7  m         n                                  r  U            x                                              !            >            [  3          x  P            m         a                                     <!         =!  s"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ?         @                                    ?         @                        	          
  ?         @                                    ?         @                                    ?         @                                    ?         @                                    ?         @               !         !  "          #  ?$         @$  %          N7  68          |8  d9          9  :          :  ;          <  <         K  L         L  .N         0N  nO         pO  P         P  Q         Q  .S         0S  nT         pT  U         U  V         V  .X         0X  nY         pY  Z         Z  [         [  .]         0]  n^         p^  _         _  `         `  .b         0b  nc         pc  d         d  e         e  .g         0g  nh         ph  i         i  j         j  .l         0l  nm         pm  n         n  o         o  .q         0q  nr           3         4  s         t                        3         4  s         t                        3         4  s         t                        3         4  s         t                        3         4  s         t                        3         4  s         t                        3         4  s         t                        3                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ;!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     *
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-<-x    <-J-|    J-}-\    }--    --h    -b-    b-qx-	    qx--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               6!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      +         ,  b	         c	  
         
             $         %  [         \                                               :  +          s  d                                              ;             X   0!          u!  M"          "  j#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            >         @  ~                                  >         @  ~                      	          
  >         @  ~                                  >         @  ~                                  >         @  ~                                  >         @  ~                                  >         @  ~             !         !  "          #  >$         @$  ~%         %  &         &  '          (  >)         @)  ~*         *  +         +  ,          -  >.         @.  ~/         /  0         0  1          2  >3         @3  ~4         4  5         5  6          7  >8         @8  ~9         9  :         :  ;          <  >=         @=  ~>         >  ?         ?  @          A  >B         @B  ~C         C  D         D  E          F  >G         @G  ~H         H  I         I  J          K  ?L         @L  M         M  N         N  O          P  ?Q         @Q  R         R  S         S  T          U  ?V         @V  W         W  X         X  Y          Z  ?[         @[  \         \  ]         ]  ^          _  ?`         @`  a         a  b         b  c          d  ?e         @e  f         f  g         g  h          i  ?j         @j  k         k  l         l  m          R  J                      ΃  Ƅ                      J  B            _         `                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               =!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 m
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-y-     y--    -6-x    6-%E-|    %E-sx-\    sx--    --h    -ob-    ob-n-	    n--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               4!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     <            Z  2          x  P                                              A	  +
          z
  d                                !  W         X                                   P         Q                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            <         >  z         |                        4         6  r         t             	         	  ,         .  j         l                        $         &  b         d                                   Z         \                                   R         T                                   J          L   !         !  "         "  $         $  B%          <  =          =  >          >  ?          @  @          BA  B         H  !J         "J  _K         `K  L         L  M         M  O         O  WP         XP  Q         Q  R         R  T         T  OU         PU  V         V  W         W  	Y         
Y  GZ         HZ  [         [  \         \  ^         ^  ?_         @_  }`         ~`  a         a  b         b  7d         8d  ue         ve  f         f  g         g  /i         0i  mj         nj  k         k  l         l  'n         o  q         q  Er         Fr  s         s  t         t  u          v  =w         >w  {x         |x  y         y  z         z  5|         6|  s}         t}  ~         ~             -         .  k         l                        %         &  c         d             ߉                      [         \             ׎         ؎             S         T             ϓ         Г                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               6!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     +
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-5-x    5-D-|    D-w-\    w-s-    s--h    -cb-    cb-kx-	    kx--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               8!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         #            F  $          i  G            j                                              $	  
          K
  )         r                                   M         N           H  ~                                 #         $  Z                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         <  6            y                                  H  B         p&  '         '  (         (  8*         9*  {+         |+  ,         ,  .         .  D/         E/  0         0  1         1  3         3  P4         Q4  5         5  6         6  8         8  \9         ]9  :         :  ;         ;  %=         &=  h>         i>  ?         ?  @         @  1B         2B  tC         uC  D         D  E         E  =G         >G  H         H  I         I  K         K  IL         JL  M         M  N         N  P         P  QQ         VQ  R         R  S         S  U         U  ]V         bV  W         W  X         X  &Z         +Z  i[         n[  \         \  ]         ]  2_         7_  u`         z`  a         a  b          c  >d         Cd  e         e  f         f  h         h  Ji         Oi  j         j  k         k  m         m  Vn         [n  o         o  p         p  r         $r  bs         gs  t         t  u         u  +w         0w  nx         sx  y         y  z         z  7|         <|  z}         }  ~         ~              C         H             Ƀ         ΃             O         T             Ո         ڈ             [         `                        $         )  g         l                        0         5  s         x                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               2!         <!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 m
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-y-     y--    -6-x    6-E-|    E-mx-\    mx--    --h    -ib-    ib-n-	    n--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               4!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     :            W  /          t  L            |                                  <	  '
          u
  `                                 R         S                                   K         L                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            =         >  {         |                        5         6  s         t             	         	  -         .  k         l                        %         &  c         d                                   [         \                                   S         T                                   K          L   !         !  "         "  $         $  C%          fB  >C          C  bD          D  E          E  F          F  G         H   J         "J  ^K         `K  L         L  M         M  O         O  VP         XP  Q         Q  R         R  T         T  NU         PU  V         V  W         W  Y         
Y  FZ         HZ  [         [  \         \   ^         ^  >_         @_  |`         ~`  a         a  b         b  6d         8d  te         ve  f         f  g         g  .i         0i  lj         nj  k         k  l         l  &n         o  q         q  Dr         Fr  s         s  t         t  u          v  <w         >w  zx         |x  y         y  z         z  4|         6|  r}         t}  ~         ~             ,         .  j         l                        $         &  b         d             މ                      Z         \             ֎         ؎             R         T             Γ         Г                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               4!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     *
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     --    -5-x    5-D-|    D-w-\    w-s-    s--h    -ib-    ib-qx-	    qx--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               6!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         #             F  #          i  F            i                                              $	  
          K
  *         r                                   M         N           H  ~                                 #         $  Z                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                         !          "  #          T#  N$          $  %         p&  '         '  (         (  7*         9*  z+         |+  ,         ,   .         .  C/         E/  0         0  1         1  3         3  O4         Q4  5         5  6         6  8         8  [9         ]9  :         :  ;         ;  $=         &=  g>         i>  ?         ?  @         @  0B         2B  sC         uC  D         D  E         E  <G         >G  H         H  I         I  K         K  HL         JL  M         M  N         N  P         P  RQ         VQ  R         R  S         S  U         U  ^V         bV  W         W  X         X  'Z         +Z  j[         n[  \         \  ]         ]  3_         7_  v`         z`  a         a  b          c  ?d         Cd  e         e  f         f  h         h  Ki         Oi  j         j  k         k  m         m  Wn         [n  o         o  p         p   r         $r  cs         gs  t         t  u         u  ,w         0w  ox         sx  y         y  z         z  8|         <|  {}         }  ~         ~             D         H             ʃ         ΃             P         T             ֈ         ڈ             \         `                        %         )  h         l                        1         5  t         x                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               4!         @!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 m
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     --    -_;-x    _;-I-|    I-G}-\    G}--    --h    -ub-    ub-Sx-	    Sx--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               8!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 S	         T	  
         
                                   )         *  `         a              |                                              4            [  ?          ~  b                          !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            :         <  v         x                        *         ,  f         h             	         	             V         X                        
           F         H                                   6         8  r         t                        &         (  b         d                                     R!         T!  "         "  #         #  %         %  B&         D&  ~'         '  (         (  )         )  2+         4+  n,         p,  -         -  .         .  "0         $0  ^1         `1  2         2  3         3  5         5  N6         P6  7         7  8         8  :         :  >;         @;  z<         |<  =         =  >         >  .@         0@  jA         lA  B         B  C         C  E          E  ZF         \F  G         G  H         H  J         J  WK         XK  L         L  M         M  O         O  GP         HP  Q         Q  R         R  S         S  7U         8U  sV         tV  W         W  X         X  'Z         (Z  c[         d[  \         \  ]         ]  _         _  S`         T`  a         a  b         b  d         d  Ce         De  f         f  g         g  h         h  3j         4j  ok         pk  l         l  m         m  #o          $r  $s          gs  gt          t  u          u  v          0w  0x                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               A!         B!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 [
  {      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-[-     [--    -;-x    ;-J-|    J-}-\    }-7-    7--h    -b-    b--	    --h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               /!         0!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      &         '  ]	         ^	  
         
                         V         W                                              5  %          n  ^                                              7             U   -!          s!  K"          "  i#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            B         C                                   N         O                        
         
  Z         [                        #         $  f         g                        /         0  r         s                        ;         <  ~                                 G         H                          "         "  S#         T#  $         $  %         %  '         '  _(         `(  )         )  *         *  (,         ),  k-         l-  .         .  /         /  41         51  w2         x2  3         3  4         4  @6         A6  7         7  8         8  	:         
:  L;         M;  <         <  =         =  ?         ?  X@         Y@  A         A  B         B  !D         "D  dE         eE  F         F  G         G  -I         .I  pJ         qJ  K          K  L          L  M          N  N          2O  P          \P  @Q         n  o         o  !q         &q  dr         	  G         L             ͙         ҙ             S         X             ٞ         ޞ           !  _         d                        (         -  k         p                        4         9  w         |                        @         E             Ʊ         ˱  	           L         Q             Ҷ         ׶                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               G!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-y-     y--    -_;-x    _;-I-|    I-G}-\    G}--    --h    -{b-    {b-Sx-	    Sx--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               6!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 S	         T	  
         
                                   )         *  `         a              }                                              4            [  >          ~  a                          !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ;         <  w         x                        +         ,  g         h             	         	             W         X                                   G         H                                   7         8  s         t                        '         (  c         d                                     S!         T!  "         "  #         #  %         %  C&         D&  '         '  (         (  )         )  3+         4+  o,         p,  -         -  .         .  #0         $0  _1         `1  2         2  3         3  5         5  O6         P6  7         7  8         8  :         :  ?;         @;  {<         |<  =         =  >         >  /@         0@  kA         lA  B         B  C         C  E          E  [F         \F  G         G  H         H  J         J  VK         XK  L         L  M         M  
O         O  FP         HP  Q         Q  R         R  S         S  6U         8U  rV         tV  W         W  X         X  &Z         (Z  b[         d[  \         \  ]         ]  _         _  R`         T`  a         a  b         b  d         d  Be         De  ~f         f  g         g  h         h  2j         4j  nk         pk  l         l  m         m  "o          sx  sy          y  z          z  {          <|  <}          }  ~                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               E!         F!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 [
  {      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-U-     U--    -;-x    ;-J-|    J-}-\    }-7-    7--h    -ub-    ub--	    --h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ,!         .!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      +         ,  b	         c	  
         
             $         %  [         \                                               :  +          s  d                                              ;             X   0!          u!  M"          "  j#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            A         C                        
           M         O                        
         
  Y         [                        "         $  e         g                        .         0  q         s                        :         <  }                                 F         H                          "         "  R#         T#  $         $  %         %  '         '  ^(         `(  )         )  *         *  ',         ),  j-         l-  .         .  /         /  31         51  v2         x2  3         3  4         4  ?6         A6  7         7  8         8  :         
:  K;         M;  <         <  =         =  ?         ?  W@         Y@  A         A  B         B   D         "D  cE         eE  F         F  G         G  ,I         .I  oJ         qJ  K          Q  jR          R  S          S  T          U  U          .V  W         n  o         o  "q         &q  er         	  H         L             Ι         ҙ             T         X             ڞ         ޞ           !  `         d                        )         -  l         p                        5         9  x         |                        A         E             Ǳ         ˱  
           M         Q             Ӷ         ׶                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               C!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-U-     U--    -y6-x    y6-E-|    E-ax-\    ax--    --h    -b-    b-Mx-	    Mx--h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     T            !          \!  1          
2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     :            W  /          t  L                                              6	  $
          m
  [                                 N         O                                   E         F  ~                                                    b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                     <  *          p  ^                                 &  '         '  )         ")  `*         e*  +         +  ,         ,  ).         ..  l/         q/  0         0  1         1  53         :3  x4         }4  5         5  6         7  A8         F8  9         9  :         :  
<         <  M=         R=  >         >  ?         ?  A         A  YB         ^B  C         C  D         D  "F         'F  eG         jG  H         H  I         I  .K         3K  qL         vL  M         M  N         N  >P         ?P  Q         Q  R         R  T         T  JU         KU  V         V  W         W  Y         Y  VZ         WZ  [         [  \         \  ^          ^  b_         c_  `         `  a         a  +c         ,c  nd         od  e         e  f         f  7h         8h  zi         {i  j         j   l         l  Cm         Dm  n         n  o         o  q         q  Or         Pr  s         s  t         t  v         v  [w         \w  x         x  y         y  ${         %{  g|         h|  }         }  ~         ~  0         1  s         t                        <         =                      Ç             H         I             Ό         ό             T         U             ڑ         ۑ             `         a                        )                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               O!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 ]
  }      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -o,-X     o,-%-     %--    -6-x    6-D-|    D-w-\    w--    --h    -Eb-    Eb-_x-	    _x--h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     M                       B!  1          1  `=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         !            B  $          c  E            f                                               	  
          I
  /         r                                   U         V           \                                   ?         @  x                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                             6  -          p  g         &  '         '  !)         ")  d*         e*  +         +  ,         ,  -.         ..  p/         q/  0         0  1         1  93         :3  |4         }4  5         5  7         7  E8         F8  9         9  :         :  <         <  Q=         R=  >         >  ?         ?  A         A  ]B         ^B  C         C  D         D  &F         'F  iG         jG  H         H  I         I  2K         3K  uL         vL  M         M  N         N  :P         ?P  }Q         Q  R         R  T         T  FU         KU  V         V  W         W  Y         Y  RZ         WZ  [         [  \         \  ^          ^  ^_         c_  `         `  a         a  'c         ,c  jd         od  e         e  f         f  3h         8h  vi         {i  j         j  k         l  ?m         Dm  n         n  o         o  q         q  Kr         Pr  s         s  t         t  v         v  Ww         \w  x         x  y         y   {         %{  c|         h|  }         }  ~         ~  ,         1  o         t                        8         =  {                    Ç             D         I             ʌ         ό             P         U             ֑         ۑ             \         a                        %                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               /!         0!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 k
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,-O-     O--    -6-x    6-E-|    E-gx-\    gx--    --h    -b-    b-Mx-	    Mx--h
    --
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     S                       X!  1          2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     <            Z  2          x  P                                              ;	  *
          r
  a                                 S         T                                   J         K                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                     @  .          t  b                                 &  '         '  )         ")  a*         e*  +         +  ,         ,  *.         ..  m/         q/  0         0  1         1  63         :3  y4         }4  5         5  6         7  B8         F8  9         9  :         :  <         <  N=         R=  >         >  ?         ?  A         A  ZB         ^B  C         C  D         D  #F         'F  fG         jG  H         H  I         I  /K         3K  rL         vL  M         M  N         N  =P         ?P  Q         Q  R         R  T         T  IU         KU  V         V  W         W  Y         Y  UZ         WZ  [         [  \         \  ^          ^  a_         c_  `         `  a         a  *c         ,c  md         od  e         e  f         f  6h         8h  yi         {i  j         j  k         l  Bm         Dm  n         n  o         o  q         q  Nr         Pr  s         s  t         t  v         v  Zw         \w  x         x  y         y  #{         %{  f|         h|  }         }  ~         ~  /         1  r         t                        ;         =  ~                    Ç             G         I             ͌         ό             S         U             ّ         ۑ             _         a                        (                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               O!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 ]
  }      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     --X     -հ-     հ-۰-    ۰-0-p    0-;-T    ;-Ed-4    Ed--    --8    -Qb-    Qb-_x-	    _x--8
    --
                   =         b      J      7   U   7   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     L                       >!  0                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 64 })  outputElements  5184    input0Elements  31104   input1Elements  24576   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     4                           #         b      J      7   M     operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       @                         !            B  #          c  D            k                                             (	  `
         a
                                              !         "  Z         [                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     4                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  1728    input0Elements  1728    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       x                                  !  "          X"  O#         &  '         '   )         ")  c*         e*  +         +  ,         ,  ,.         ..  o/         q/  0         0  1         1  83         :3  {4         }4  5         5  7         7  D8         F8  9         9  :         :  <         <  P=         R=  >         >  ?         ?  A         A  \B         ^B  C         C  D         D  %F         'F  hG         jG  H         H  I         I  1K         3K  tL         vL  M         M  N         N  ;P         ?P  ~Q         Q  R         R  T         T  GU         KU  V         V  W         W  Y         Y  SZ         WZ  [         [  \         \  ^          ^  __         c_  `         `  a         a  (c         ,c  kd         od  e         e  f         f  4h         8h  wi         {i  j         j  k         l  @m         Dm  n         n  o         o  	q         q  Lr         Pr  s         s  t         t  v         v  Xw         \w  x         x  y         y  !{         %{  d|         h|  }         }  ~         ~  -         1  p         t                        9         =  |                    Ç             E         I             ˌ         ό             Q         U             ב         ۑ             ]         a                        &                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               3!         4!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 k
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,-I-     I--    -e;-x    e;-I-|    I-M}-\    M}-9-    9--h    -ob-    ob-ǂ-	    ǂ--h
    -%-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     P                       T!  1          1  ~=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 O	         P	  
         
                                    9         :  r         s                                                '            P  ,          y  Q            r                          !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            >         C                                   J         O                        
         
  V         [                                 $  b         g                        +         0  n         s                        7         <  z                                  C         H                          "         "  O#         T#  $         $  %         %  '         '  [(         `(  )         )  *         *  $,         ),  g-         l-  .         .  /         /  01         51  s2         x2  3         3  4         4  <6         A6  7         7  8         8  :         
:  H;         M;  <         <  =         =  ?         ?  T@         Y@  A         A  B         B  D         "D  `E         eE  F         F  G         G  )I         .I  lJ         qJ  K          *            b  P                      ҋ            
           Z             ߖ           "         #  e         f                        .         /  q         r                        :         ;  }         ~                        F         G             ̩         ͩ             R         S             خ         ٮ             ^         _                        '         (  j         k                        3                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               >!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -i,-X     i,--     --    -;-x    ;-J-|    J-}-\    }-=-    =--P    -cb-    cb-c-	    c--D
    -%-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     M                       <!  1          1  W=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 @	         A	  y
         z
                        7         8  p         q                                               Q  5            l                                              D   !          k!  C"          "  j#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            9         :  s         t                        !         "  [         \             	         	  	         
  C         D  }         ~                        +         ,  e         f                                   M         N                                   5         6  o         p                        !         !  W"         X"  #         #  $         $  &         &  ?'         @'  y(         z(  )         )  *         *  ',         (,  a-         b-  .         .  /         /  1         1  I2         J2  3         3  4         4  5         5  17         27  k8         l8  9         9  :         :  <         <  S=         T=  >         >  ?         ?  A         A  ;B         <B  uC         vC  D         D  E         E  #G         $G  \H          ^H  JI          I  ~J          J  K          K  L          .M  N         vl  m         m  n         n  "p         $p  \q         ^q  r         r  s         s  
u         u  Dv         Fv  ~w         w  x         x  y         y  ,{         .{  f|         h|  }         }  ~         ~             N         P                      ă             6         8  p         r                                    X         Z             ̎                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               6!         @!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 [
  {                   
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,-I-     I--    -k;-x    k;-J-|    J-S}-\    S}-9-    9--h    -ob-    ob-ǂ-	    ǂ--h
    -%-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     Q                       T!  1          1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 O	         P	  
         
                                    9         :  r         s                                                '            P  -          y  Q            s                          !          "  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ?         C                                   K         O                        
         
  W         [                                  $  c         g                        ,         0  o         s                        8         <  {                                 D         H                          "         "  P#         T#  $         $  %         %  '         '  \(         `(  )         )  *         *  %,         ),  h-         l-  .         .  /         /  11         51  t2         x2  3         3  4         4  =6         A6  7         7  8         8  :         
:  I;         M;  <         <  =         =  ?         ?  U@         Y@  A         A  B         B  D         "D  aE         eE  F         F  G         G  *I         .I  mJ         qJ  K          B  0          z  h                        ؒ          "           Z             ޖ           !         #  d         f                        -         /  p         r                        9         ;  |         ~                        E         G             ˩         ͩ             Q         S             ׮         ٮ             ]         _                        &         (  i         k                        2                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               >!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -i,-X     i,--     --    -;-x    ;-J-|    J-}-\    }-C-    C--P    -cb-    cb-c-	    c--D
    -%-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     L                       :!  1          1  S=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 ;	         <	  t
         u
                        2         3  k         l                                               L  1            h                                              A   !          i!  A"          "  i#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            8         :  r         t                                  "  Z         \             	         	           
  B         D  |         ~                        *         ,  d         f                                   L         N                                   4         6  n         p                        !         !  V"         X"  #         #  $         $  &         &  >'         @'  x(         z(  )         )  *         *  &,         (,  `-         b-  .         .  /         /  1         1  H2         J2  3         3  4         4  5         5  07         27  j8         l8  9         9  :         :  <         <  R=         T=  >         >  ?         ?   A         A  :B         <B  tC         vC  D         D  E         E  "G         $G  ]H          bN  NO          O  P          P  Q          Q  R          2S  T         vl  m         m  n         n  #p         $p  ]q         ^q  r         r  s         s  u         u  Ev         Fv  w         w  x         x  y         y  -{         .{  g|         h|  }         }  ~         ~             O         P             Ã         ă             7         8  q         r                                    Y         Z             ͎                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               8!         @!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 [
  {                   
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,--     -)-    )-;-x    ;-I-|    I-|-\    |--    -e-h    e-b-    b-ǂ-	    ǂ--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               =!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             8         9  q         r                                  |  T            u                                               !            >            [  3          x  P            m         W                                     :!         ;!  s"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            B         C                                   N         O                        
         
  Z         [                        #         $  f         g                        /         0  r         s                        ;         <  ~                                 G         H                          "         "  S#         T#  $         $  %         %  '         '  [(         `(  )         )  *         *  $,         ),  g-         l-  .         .  /         /  01         51  s2         x2  3         3  4         4  <6         A6  7         7  8         8  :         
:  H;         M;  <         <  =         =  ?         ?  T@         Y@  A         A  B         B  D         "D  `E         eE  F         F  G         G  )I          I  oJ          J  K          K  L          M  M          (N  O         l  m         m  o         o  Rp         Sp  q         q  r         r  t         t  ^u         _u  v         v  w         w  'y         (y  jz         kz  {         {  |         |  3~         4~  v         w                        ?         @             Ņ         ƅ           	  K         L             ъ         Ҋ             W         X             ݏ         ޏ            !  c           ؔ         ٔ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     E                       2!  w1          1  @=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,-s-     s--    -;;-    ;;-I-    I-#}-d    #}--    --p    -3b-    3b-W-
    W-e-
    e-ð-t    ð-3׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               5!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             8         9  q         r                                                                      7            ^  6            g                                              	                                    8          9   q!         r!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                 $            g  b                     T#  $         $  %         %  '         '  C(         D(  )         )  *         *  +         +  3-         4-  o.         p.  /         /  0         0  #2         $2  _3         `3  4         4  5         5  7         7  O8         P8  9         9  :         :  <         <  ?=         @=  {>         |>  ?         ?  @         @  /B         0B  kC         lC  D         D  E         E  G          G  [H         \H  I         J  K         K  8M         :M  tN         vN  O         O  P         P  (R         *R  dS         fS  T         T  U         U  W         W  TX         VX  Y         Y  Z         Z  \         
\  D]         F]  ^         ^  _         _  `         `  4b         6b  pc         rc  d         d  e         e  $g         &g  `h         bh  i         i  j         j  l         l  Pm         Rm  n         n  o         o  q         q  @r         Br  |s         ~s  t         t  u         u  0w         2w  lx         nx  y         y  z         z   |         "|  \}         ^}  ~         ~                        L         N             Ą         Ƅ              <         >  x         z                        ,         .  h         j                                   X         Z                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     M                       N!  1          1  r=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
                   
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                      
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-;-x    ;-I-|    I-|-\    |--    -k-h    k-	b-    	b-ǂ-	    ǂ--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ;!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             8         9  q         r                                  |  T            v                                              &            D            b  :            X            v         a                                     D!         E!  }"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            A         C                        
           M         O                        
         
  Y         [                        "         $  e         g                        .         0  q         s                        :         <  }                                 F         H                          "         "  R#         T#  $         $  %         %  '         '  \(         `(  )         )  *         *  %,         ),  h-         l-  .         .  /         /  11         51  t2         x2  3         3  4         4  =6         A6  7         7  8         8  :         
:  I;         M;  <         <  =         =  ?         ?  U@         Y@  A         A  B         B  D         "D  aE         eE  F         F  G         G  *I          MO  (P          rP  MQ          Q  rR          R  S          S  T         l  m         m  o         o  Qp         Sp  q         q  r         r  t         t  ]u         _u  v         v  w         w  &y         (y  iz         kz  {         {  |         |  2~         4~  u         w                        >         @             ą         ƅ           	  J         L             Њ         Ҋ             V         X             ܏         ޏ           !  b           ה         ٔ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     F                       6!  |1          1  G=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,-y-     y--    -G;-    G;-I-    I-/}-d    /}--    --p    -3b-    3b-W-
    W-e-
    e-ð-t    ð-3׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               7!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             8         9  q         r                                                                      :            b  :            k                                                                      	         
  B          C   {!         |!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                     0  +          s  n                                 T#  $         $  %         %  '         '  B(         D(  ~)         )  *         *  +         +  2-         4-  n.         p.  /         /  0         0  "2         $2  ^3         `3  4         4  5         5  7         7  N8         P8  9         9  :         :  <         <  >=         @=  z>         |>  ?         ?  @         @  .B         0B  jC         lC  D         D  E         E  G          G  ZH         \H  I         J  K         K  9M         :M  uN         vN  O         O  P         P  )R         *R  eS         fS  T         T  U         U  W         W  UX         VX  Y         Y  Z         Z  	\         
\  E]         F]  ^         ^  _         _  `         `  5b         6b  qc         rc  d         d  e         e  %g         &g  ah         bh  i         i  j         j  l         l  Qm         Rm  n         n  o         o  q         q  Ar         Br  }s         ~s  t         t  u         u  1w         2w  mx         nx  y         y  z         z  !|         "|  ]}         ^}  ~         ~                        M         N             ń         Ƅ             =         >  y         z                        -         .  i         j                                   Y         Z                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     L                       N!  1          1  q=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
                   
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                      
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-k;-x    k;-J-|    J-S}-\    S}--    --P    -9b-    9b-c-	    c-1-D
    1-9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               :!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 O	         P	  
         
                                    9         :  r         s                                                '            P  ,          y  Q            r                          !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            >         C                                   J         O                        
         
  V         [                                 $  b         g                        +         0  n         s                        7         <  z                                  C         H                          "         "  O#         T#  $         $  %         %  '         '  [(         `(  )         )  *         *  $,         ),  g-         l-  .         .  /         /  01         51  s2         x2  3         3  4         4  <6         A6  7         7  8         8  :         
:  H;         M;  <         <  =         =  ?         ?  T@         Y@  A         A  B         B  D         ng  h         h  i         i  2k         7k  ul         zl  m         m  n          o  >p         Cp  q         q  r         r  t         t  Ju         Ou  v         v  w         w  y         y  Sz         Xz  {         {  |         |  ~         ~  ]         b                        &         +  l         m                        2         3  t         u                        =         >           4  r         w                          Ϝ                      D  /          t  _                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     K                       P!  1          1  s=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 i
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    <     -,-p     ,--     --    -6-    6-D-    D-w-t    w-G4-     G4-;-X    ;--    -ub-p	    ub-]-x
    ]-k-
    k--    -ֳ-\    ֳ--<                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               7!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         !            B  $          c  E            f                                               	  
          I
  /         r                                   U         V           \                                   ?         @  x                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        (#  c$         CF  ~G         G  H         H  I         I  8K         :K  vL         xL  M         M  N         N  0P         2P  nQ         pQ  R         R  S         S  (U         *U  fV         hV  W         W  X         X   Z         "Z  ^[         `[  \         \  ]         ]  _         _  V`         X`  a         a  b         b  d         d  Ne         Pe  f         f  g         g  i         
i  Fj         Hj  k         k  l         l   n         n  >o         @o  |p         ~p  q         q  r         r  6t         8t  tu         vu  v         v  w         w  .y         0y  lz         nz  {         {  |         |  &~         (~  d         f                                    \         ^             ؆         چ             T         V             Ћ         ҋ             L         N             Ȑ         ʐ             C         E                                   9         ;  w         y                        1         3  o         q                        )         +  g         i                        !         #  _          a  Z                      ݪ  ֫                      Y  R           ӯ         կ             O         Q             ˴                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               8!         H!  1         1  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
                                           b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            0     -,-X     ,--     -#-    #-q;-x    q;-J-|    J-Y}-\    Y}-%-    %--P    -9b-    9b-c-	    c-1-D
    1-9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               9!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 O	         P	  
         
                                    9         :  r         s                                                '            P  -          y  Q            s                          !          "  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            ?         C                                   K         O                        
         
  W         [                                  $  c         g                        ,         0  o         s                        8         <  {                                 D         H                          "         "  P#         T#  $         $  %         %  '         '  \(         `(  )         )  *         *  %,         ),  h-         l-  .         .  /         /  11         51  t2         x2  3         3  4         4  =6         A6  7         7  8         8  :         
:  I;         M;  <         <  =         =  ?         ?  U@         Y@  A         A  B         B  D         ng  h         h  i         i  3k         7k  vl         zl  m         m  n          o  ?p         Cp  q         q  r         r  t         t  Ku         Ou  v         v  w         w  y         y  Tz         Xz  {         {  |         |  ~         ~  ^         b                        '         +  k         m                        1         3  s         u                        <         >           4  s         w                        ԡ                        4            d  O                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     L                       P!  1          1  t=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 j
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    <     -,-p     ,--     -#-    #-6-    6-D-    D-w-t    w-4-     4-;-X    ;--    -ub-p	    ub-]-x
    ]-k-
    k--    -ֳ-\    ֳ--<                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               9!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         !            B  #          c  D            e                                               	  
          I
  0         r                                   U         V           \                                   ?         @  x                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        (#  d$         CF  G         G  H         H  I         I  9K         :K  wL         xL  M         M  N         N  1P         2P  oQ         pQ  R         R  S         S  )U         *U  gV         hV  W         W  X         X  !Z         "Z  _[         `[  \         \  ]         ]  _         _  W`         X`  a         a  b         b  d         d  Oe         Pe  f         f  g         g  	i         
i  Gj         Hj  k         k  l         l  n         n  ?o         @o  }p         ~p  q         q  r         r  7t         8t  uu         vu  v         v  w         w  /y         0y  mz         nz  {         {  |         |  '~         (~  e         f                                    ]         ^             ن         چ             U         V             ы         ҋ             M         N             ɐ         ʐ             D         E                                   :         ;  x         y                        2         3  p         q                        *         +  h         i                        "         #  `         a             ܪ         ݪ             X         Y                        կ  ΰ                      Q  J                                          b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               8!         H!  1         1  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
                                           b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            8     -,-h     ,--     -#-    #-;-    ;-I-    I-|-l    |--    --P    --    -b-h	    b-c-p
    c-q-
    q-۰-    ۰-9׳-L                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               7!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             8         9  q         r                                  |  T            u                                               !            >            [  3          x  P            m         W                                     :!         ;!  s"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        d#  $         $  %         %  "'         $'  b(         d(  )         )  *         *  ",         $,  b-         d-  .         .  /         /  "1         $1  b2         d2  3         3  4         4  "6         $6  b7         d7  8         8  9         9  ";         $;  b<         d<  =         =  >         >  "@         $@  bA         dA  B         B  C         C  "E         $E  bF         dF  G         G  H         H  "J         $J  cK         dK  L         L  M         M  #O         $O  cP         dP  Q         Q  R         R  #T         $T  cU         dU  V         V  W         W  #Y         $Y  cZ         dZ  [         [  \         \  #^         $^  c_         d_  `         `  a         a  #c         $c  cd         dd  e         e  f         f  #h         $h  ci         di  j         j  k         k  #m         $m  cn         dn  o         ds  t         t  u         u  "w         $w  bx         dx  y         y  z         z  "|         $|  b}         d}  ~         ~             "         $  b         d                        "         $  b         d                        "          $            d  _                        ߏ          $           d                        "         $  b         d                                b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               C!         ^!  1         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
                                           b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     2
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-G;-x    G;-I-|    I-/}-\    /}-2-    2--h    -b-    b-n-	    n--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               8!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             8         9  q         r                                                                      7            ^  6            g                                              	                                    8          9   q!         r!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ?         @                                    ?         @                        	          
  ?         @                                    ?         @                                    ?         @                                    ?         @                                    ?         @               !          !  "          "  #          #  $          %  %          <&  '         F  H         H  GI         I  J         J  L         L  QM         SM  N         N  O         O  Q         Q  QR         SR  S         S  T         T  V         V  QW         SW  X         X  Y         Y  [         [  Q\         S\  ]         ]  ^         ^  `         `  Qa         Sa  b         b  c         c  e         e  Qf         Sf  g         g  h         h  j         j  Qk         Sk  l         l  m         m  o         o  p         p  #r         $r  cs           א         ؐ             W         X             ו         ؕ             W         X             ך         ؚ             W         X             ן         ؟             W         X             פ         ؤ             W         X             ש         ة             W         X             ׮         خ             W         X                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ;!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     5
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     -#-    #-;-    ;-I-    I-|-l    |--    -{-P    {--    -b-h	    b-c-p
    c-q-
    q-۰-    ۰-9׳-L                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               7!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             8         9  q         r                                  |  T            v                                              &            D            b  :            X            v         a                                     D!         E!  }"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        d#  $         $  %         %  #'         $'  c(         d(  )         )  *         *  #,         $,  c-         d-  .         .  /         /  #1         $1  c2         d2  3         3  4         4  #6         $6  c7         d7  8         8  9         9  #;         $;  c<         d<  =         =  >         >  #@         $@  cA         dA  B         B  C         C  #E         $E  cF         dF  G         G  H         H  #J         $J  bK         dK  L         L  M         M  "O         $O  bP         dP  Q         Q  R         R  "T         $T  bU         dU  V         V  W         W  "Y         $Y  bZ         dZ  [         [  \         \  "^         $^  b_         d_  `         `  a         a  "c         $c  bd         dd  e         e  f         f  "h         $h  bi         di  j         j  k         k  "m         $m  bn         dn  o         ds  t         t  u         u  #w         $w  cx         dx  y         y  z         z  #|         $|  c}         d}  ~         ~             #         $  c         d                        #         $  c         d                        #         $  c         d                        #         $  c          d  _                        ߔ          $            d  _                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               C!         ^!  1         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
                                           b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     1
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-M;-x    M;-I-|    I-5}-\    5}-2-    2--h    -b-    b-n-	    n--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               8!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             8         9  q         r                                                                      :            b  :            k                                                                      	         
  B          C   {!         |!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            >         @  ~                                  >         @  ~                      	          
  >         @  ~                                  >         @  ~                                  >         @  ~                                  >         @  ~                                  >         @  ~             !          ['  5(          z(  T)          )  s*          *  +          +  ,         F  H         H  FI         I  J         J  L         L  RM         SM  N         N  O         O  Q         Q  RR         SR  S         S  T         T  V         V  RW         SW  X         X  Y         Y  [         [  R\         S\  ]         ]  ^         ^  `         `  Ra         Sa  b         b  c         c  e         e  Rf         Sf  g         g  h         h  j         j  Rk         Sk  l         l  m         m  o         o  p         p  "r         $r  bs           ֐         ؐ             V         X             ֕         ؕ             V         X             ֚         ؚ             V         X             ֟         ؟             V         X             ֤         ؤ             V         X             ֩         ة             V         X             ֮         خ             V         X                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               =!         R!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     6
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     -)-    )-6-    6-+E-    +E-yx-l    yx-8-    8--x    -b-	    b-c-
    c-q-
    q--|    -ֳ-    ֳ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               9!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     :            W  /          t  L                                              6	  $
          m
  [                                 N         O                                   E         F  ~                                                    b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         %  &          
'  '          :(  ")          j)  R*          *  +         zI  J         J  K         K  2M         4M  pN         rN  O         O  P         P  *R         ,R  hS         jS  T         T  U         U  "W         $W  `X         bX  Y         Y  Z         Z  \         \  X]         Z]  ^         ^  _         _  a         a  Pb         Rb  c         c  d         d  
f         f  Hg         Jg  h         h  i         i  k         k  Al         Bl  m         m  n         n  o         o  9q         :q  wr         xr  s         s  t         t  1v         2v  ow         pw  x         x  y         y  ){         *{  g|         h|  }         }  ~         ~  !         "  _         `             ۃ         ܃             W         X             ӈ         Ԉ             O         P             ˍ         ̍  	         
  G         H             Ò         Ē             ?         @  }         ~                        7         8  u         v                        /         0  m         n                        '         (  e         f                                    ]         ^             ٬         ڬ             U         V             ѱ         ұ             L         N             ȶ         ޶             Y                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               :!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
                   
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     n
            0     -,-X     ,--     -#-    #-6-x    6-D-|    D-x-\    x--    --h    -ub-    ub-n-	    n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               6!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         !            B  $          c  E            f                                               	  
          I
  /         r                                   U         V           \                                   ?         @  x                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            B         C                                   N         O                        
         
  Z         [                        #         $  f         g                        /         0  r         s                        ;         <  ~                                 G         H                          "         "  S#         T#  $         $  %         %  '         '  [(         `(  )         )  *         *  $,         ),  g-         l-  .         .  /         /  01         51  s2         x2  3         3  4         4  <6         A6  7         7  8         8  :         
:  H;         M;  <         <  =         =  ?         ?  T@         Y@  A         A  B         B  D         "D  `E         eE  F          nR  XS          S  T          T  U          U  V          W   X         i  j         j  7l         <l  zm         m  n         n   p         p  Cq         Hq  r         r  s         s  u         u  Ov         Tv  w         w  x         x  z         z  [{         `{  |         |  }         }  $         )  g         l                        0         5  s         x                        <         A                      ǋ           
  H         M             ΐ         Ӑ             T         Y                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               6!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     5
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     -)-    )-6-    6-1E-    1E-x-l    x-8-    8--x    -{b-	    {b-c-
    c-q-
    q--|    -ֳ-    ֳ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               9!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     <            Z  2          x  P                                              ;	  *
          r
  a                                 S         T                                   J         K                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         +  ,          ,  -          *.  /          Z/  B0          0  r1         zI  J         J  K         K  3M         4M  qN         rN  O         O  P         P  +R         ,R  iS         jS  T         T  U         U  #W         $W  aX         bX  Y         Y  Z         Z  \         \  Y]         Z]  ^         ^  _         _  a         a  Qb         Rb  c         c  d         d  f         f  Ig         Jg  h         h  i         i  k         k  @l         Bl  ~m         m  n         n  o         o  8q         :q  vr         xr  s         s  t         t  0v         2v  nw         pw  x         x  y         y  ({         *{  f|         h|  }         }  ~         ~            "  ^         `             ڃ         ܃             V         X             ҈         Ԉ             N         P             ʍ         ̍           
  F         H                      Ē              >         @  |         ~                        6         8  t         v                        .         0  l         n                        &         (  d         f                                    \         ^             ج         ڬ             T         V             б         ұ             M         N             ɶ         ޶             X                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               8!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
                   
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     m
            0     -,-X     ,--     -#-    #-6-x    6-D-|    D-x-\    x--    --h    -ub-    ub-n-	    n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               6!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         !            B  #          c  D            e                                               	  
          I
  0         r                                   U         V           \                                   ?         @  x                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            A         C                        
           M         O                        
         
  Y         [                        "         $  e         g                        .         0  q         s                        :         <  }                                 F         H                          "         "  R#         T#  $         $  %         %  '         '  \(         `(  )         )  *         *  %,         ),  h-         l-  .         .  /         /  11         51  t2         x2  3         3  4         4  =6         A6  7         7  8         8  :         
:  I;         M;  <         <  =         =  ?         ?  U@         Y@  A         A  B         B  D         "D  aE         eE  F          @X  *Y          jY  TZ          Z  ~[          [  \          \  ]         i  j         j  8l         <l  {m         m  n         n  p         p  Dq         Hq  r         r  s         s  u         u  Pv         Tv  w         w  x         x  z         z  \{         `{  |         |  }         }  %         )  h         l                        1         5  t         x                        =         A             Ë         ǋ           
  I         M             ϐ         Ӑ             U         Y                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               6!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     6
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -;-    ;-;-x    ;-I-|    I-|-\    |-'9-    '9--h    -b-    b-n-	    n--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               >!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             8         9  q         r                                  |  T            u                                               !            >            [  3          x  P            m         W                                     :!         ;!  s"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            :         <  v         x                        *         ,  f         h             	         	             V         X                        
           F         H                                   6         8  r         t                        &         (  b         d                                     R!         T!  "         "  #         #  %         %  C&         D&  '         '  (         (  )         )  3+         4+  o,         p,  -         -  .         .  #0         $0  _1         `1  2         2  3         3  5         5  O6         P6  7         7  8         8  :         :  ?;         @;  {<         |<  =         =  >         >  /@         0@  kA         lA  B         B  C         C  E          E  [F         \F  G          +S  T          KT  #U          kU  CV          V  cW          W  X         m  o         o  Op           ֕         ؕ             N         P             ƚ         Ț             >         @  z         |                        .         0  j         l                                    Z         \             Ҫ         Ԫ             J         L             ¯         į              :         <  v         x                        *         ,  f         h                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               A!         V!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     +
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-<-x    <-J-|    J-}-\    }-c6-    c6--h    -{b-    {b-wx-	    wx--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               :!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 @	         A	  y
         z
                        7         8  p         q                                               Q  5            l                                              D   !          k!  C"          "  j#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            B         C                                   N         O                        
         
  Z         [                        #         $  f         g                        /         0  r         s                        ;         <  ~                                 G         H                          "         "  S#         T#  $         $  %          ?  @          Y@  IA          A  B          B  C          "D  E         F  G         G  -I         .I  pJ         qJ  K         K  L         L  9N         :N  |O         }O  P         P  R         R  ES         FS  T         T  U         U  W         W  QX         RX  Y         Y  Z         Z  \         \  ]]         ^]  ^         ^  _         _  &a         'a  ib         jb  c         c  d         d  2f         3f  ug         vg  h         h  i         i  >k         ?k  l         l  m         m  o         o  Fp         Kp  q         q  r         r  t         t  Ru         Wu  v         v  w         w  y          y  ^z         cz  {         {  |         |  '~         ,~  j         o                        3         8  v         {                        ?         D             Ŋ         ʊ             K         ұ             S         X             ٶ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               7!         F!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 k
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -;-    ;-#;-x    #;-I-|    I-}-\    }--9-    -9--h    -b-    b-n-	    n--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               @!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             8         9  q         r                                  |  T            v                                              &            D            b  :            X            v         a                                     D!         E!  }"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ;         <  w         x                        +         ,  g         h             	         	             W         X                                   G         H                                   7         8  s         t                        '         (  c         d                                     S!         T!  "         "  #         #  %         %  B&         D&  ~'         '  (         (  )         )  2+         4+  n,         p,  -         -  .         .  "0         $0  ^1         `1  2         2  3         3  5         5  N6         P6  7         7  8         8  :         :  >;         @;  z<         |<  =         =  >         >  .@         0@  jA         lA  B         B  C         C  E          E  ZF         \F  G          X  Y          Y  Z          [  [          +\  ]          K]  #^         m  o         o  Np           ו         ؕ             O         P             ǚ         Ț             ?         @  {         |                        /         0  k         l                                    [         \             Ӫ         Ԫ             K         L             ï         į              ;         <  w         x                        +         ,  g         h                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               A!         V!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     *
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-<-x    <-J-|    J-}-\    }-c6-    c6--h    -b-    b-wx-	    wx--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               6!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 ;	         <	  t
         u
                        2         3  k         l                                               L  1            h                                              A   !          i!  A"          "  i#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            A         C                        
           M         O                        
         
  Y         [                        "         $  e         g                        .         0  q         s                        :         <  }                                 F         H                          "         "  R#         T#  $         $  %          eE  UF         F  G         G  ,I         .I  oJ         qJ  K         K  L         L  8N         :N  {O         }O  P         P  R         R  DS         FS  T         T  U         U  W         W  PX         RX  Y         Y  Z         Z  \         \  \]         ^]  ^         ^  _         _  %a         'a  hb         jb  c         c  d         d  1f         3f  tg         vg  h         h  i         i  =k         ?k  l         l  m         m  o         o  Gp         Kp  q         q  r         r  t         t  Su         Wu  v         v  w         w  y          y  _z         cz  {         {  |         |  (~         ,~  k         o                        4         8  w         {                        @         D             Ɗ         ʊ  	           L                      Ӑ  Ñ                      Y  I         ұ             T         X             ڶ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               9!         J!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 k
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -A-    A-6-x    6-CE-|    CE-x-\    x-.-    .--h    -b-    b-n-	    n--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               @!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                      @            `  8            X                                    	          E	  8
          |
  o                      $         %  ]         ^                                   T         U                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            8         :  r         t                                  "  Z         \             	         	           
  B         D  |         ~                        *         ,  d         f                                   L         N                                   4         6  n         p                        !         !  V"         X"  #         #  $         H  J         J  WK         XK  L         L  M         M  O         O  ?P         @P  yQ         zQ  R         R  S         S  'U         (U  aV         bV  W         W  X         X  Z         Z  I[         J[  \         \  ]         ]  ^         ^  1`         2`  ka         la  b         b  c         c  e         e  Sf         Tf  g         g  h         h  j         j  ;k         <k  ul         vl  m         No  p         p  q         q  r         r  5t         6t  ou         pu  v         v  w         w  y         y  Wz         Xz  {         {  |         |  ~         ~  ?         @  y         z                        '         (  a         b             Շ         և             I         J                                   1         2  k         l             ߒ                       ʪ                      8            ^  <            b                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               D!         T!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     +
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-6-x    6-D-|    D-x-\    x--    --h    -ob-    ob-wx-	    wx--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               <!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         !            B            c  @            a                                               	  
          I
  *         r                                   U         V           \                                   ?         @  x                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                        6  $          l  Z                                 %  6'         7'  y(         z(  )         )  *          +  B,         C,  -         -  .         .  0         0  N1         O1  2         2  3         3  5         5  Z6         [6  7         7  8         8  #:         $:  f;         g;  <         <  =         =  /?         0?  r@         s@  A         A  B         B  ;D         <D  ~E         E  F         F  H         H  GI         HI  J         J  K         K  M         M  ON         TN  O         O  P         P  R         R  [S         `S  T         T  U         U  $W         )W  gX         lX  Y         Y  Z         Z  0\         5\  s]         x]  ^         ^  _         _  <a         Aa  b         b  c         c  e         
e  Hf         Mf  g         g  h         h  j         j  Tk         Yk  l         l  m         m  o         "o  `p         ep  q         q  r         r  )t         .t  lu         qu  v         v  w         w  5y         :y  xz         }z  {         {  |         }  A~         F~             ǀ         ̀  
           M         R             Ӆ         ؅             Y         ^             ߊ           "         '  e         j                        .         3  q         v                        :         ?  }                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               0!         8!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 k
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -A-    A-6-x    6-=E-|    =E-x-\    x-4-    4--h    -b-    b-n-	    n--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               B!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     >            ]  5          |  T                                  	            @	  2
          w
  i                                  X         Y                                   O         P                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            9         :  s         t                        !         "  [         \             	         	  	         
  C         D  }         ~                        +         ,  e         f                                   M         N                                   5         6  o         p                        !         !  W"         X"  #         #  $         H  J         J  VK         XK  L         L  M         M  O         O  >P         @P  xQ         zQ  R         R  S         S  &U         (U  `V         bV  W         W  X         X  Z         Z  H[         J[  \         \  ]         ]  ^         ^  0`         2`  ja         la  b         b  c         c  e         e  Rf         Tf  g         g  h         h   j         j  :k         <k  tl         vl  m         No  p         p  q         q  r         r  4t         6t  nu         pu  v         v  w         w  y         y  Vz         Xz  {         {  |         |  ~         ~  >         @  x         z                        &         (  `         b             ԇ         և             H         J                                   0         2  j         l             ޒ                                 а              Բ                      B                      >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               B!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     *
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-6-x    6-D-|    D-x-\    x--    --h    -ub-    ub-wx-	    wx--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               8!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         !            B            c  ?            `                                               	  
          I
  +         r                                   U         V           \                                   ?         @  x                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                     D  2          z  h	          	  
          
           %  5'         7'  x(         z(  )         )  *          +  A,         C,  -         -  .         .  
0         0  M1         O1  2         2  3         3  5         5  Y6         [6  7         7  8         8  ":         $:  e;         g;  <         <  =         =  .?         0?  q@         s@  A         A  B         B  :D         <D  }E         E  F         F  H         H  FI         HI  J         J  K         K  M         M  PN         TN  O         O  P         P  R         R  \S         `S  T         T  U         U  %W         )W  hX         lX  Y         Y  Z         Z  1\         5\  t]         x]  ^         ^  _         _  =a         Aa  b         b  c         c  e         
e  If         Mf  g         g  h         h  j         j  Uk         Yk  l         l  m         m  o         "o  ap         ep  q         q  r         r  *t         .t  mu         qu  v         v  w         w  6y         :y  yz         }z  {         {  |         }  B~         F~             Ȁ         ̀             N         R             ԅ         ؅             Z         ^                        #         '  f         j                        /         3  r         v                        ;         ?  ~                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               2!         <!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 k
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     ---X     ---     -e-    e-;-x    ;-kJ-|    kJ-}-\    }-8-    8--h    -ub-    ub-傳-	    傳--h
    -Q׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               W!         X!  2         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 O	         P	  
         
                                    9         :  r         s                                                <            l  D            t                                    !  !          L"  $#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            >         C                                   J         O                        
         
  V         [                                 $  b         g                        +         0  n         s                        7         <  z                                  C         H                          "         "  O#         T#  $         $  %         %  '         '  [(         `(  )         )  *         *  $,         ),  g-         l-  .         .  /         /  01         51  s2         x2  3         3  4         4  <6         A6  7         7  8         8  :         
:  H;         M;  <         <  =         =  ?         ?  T@         Y@  A         A  B         B  D         "D  `E         eE  F         F  G         G  )I         .I  lJ         qJ  K          "o  p          Vp  5q          q  ir          r  s          s  t         :  |         }                        E         F             ˚         ̚             Q         R             ן         ؟             ]         ^                        &         '  i         j                        2         3  u         v                        >         ?             Ĳ         Ų             J         K             з         ѷ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               0!         6!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     --    -=<-x    =<-J-|    J-%~-\    %~-8-    8--h    -b-    b-߂-	    ߂--h
    -Q׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               2!         4!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 @	         A	  y
         z
                        7         8  p         q                                               Q  4            k                                  &            V   .!          !  ^"          "  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            B         C                                   N         O                        
         
  Z         [                        #         $  f         g                        /         0  r         s                        ;         <  ~                                 G         H                          "         "  S#         T#  $         $  %         %  '         '  _(         `(  )         )  *         *  (,         ),  k-         l-  .         .  /         /  41         51  w2         x2  3         3  4         4  @6         A6  7         7  8         8  	:         
:  L;         M;  <         <  =         =  ?         ?  X@         Y@  A         A  B         B  !D         "D  dE         eE  F         F  G         G  -I         .I  pJ         qJ  K          0W  X          VX  4Y          |Y  ZZ          Z  [          [  \         :  x         }                        A         F             ǚ         ̚  
           M         R             ӟ         ؟             Y         ^             ߤ           "         '  e         j                        .         3  q         v                        :         ?  }                    Ų             F         K             ̷         ѷ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               6!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     ---X     ---     -_-    _-;-x    ;-kJ-|    kJ-}-\    }-8-    8--h    -{b-    {b-傳-	    傳--h
    -Q׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               S!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 O	         P	  
         
                                    9         :  r         s                                    
            9            h  @            o                                   !  !          K"  ##                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ?         C                                   K         O                        
         
  W         [                                  $  c         g                        ,         0  o         s                        8         <  {                                 D         H                          "         "  P#         T#  $         $  %         %  '         '  \(         `(  )         )  *         *  %,         ),  h-         l-  .         .  /         /  11         51  t2         x2  3         3  4         4  =6         A6  7         7  8         8  :         
:  I;         M;  <         <  =         =  ?         ?  U@         Y@  A         A  B         B  D         "D  aE         eE  F         F  G         G  *I         .I  mJ         qJ  K          &u  v          Zv  9w          w  mx          x  y          y  z         :  {         }                        D         F             ʚ         ̚             P         R             ֟         ؟             \         ^                        %         '  h         j                        1         3  t         v                        =         ?             ò         Ų             I         K             Ϸ         ѷ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               2!         :!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     --    -7<-x    7<-J-|    J-~-\    ~-8-    8--h    -b-    b-傳-	    傳--h
    -Q׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               1!         2!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 ;	         <	  t
         u
                        2         3  k         l                                               L  .            e                                               O   '!          ~!  V"          "  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            A         C                        
           M         O                        
         
  Y         [                        "         $  e         g                        .         0  q         s                        :         <  }                                 F         H                          "         "  R#         T#  $         $  %         %  '         '  ^(         `(  )         )  *         *  ',         ),  j-         l-  .         .  /         /  31         51  v2         x2  3         3  4         4  ?6         A6  7         7  8         8  :         
:  K;         M;  <         <  =         =  ?         ?  W@         Y@  A         A  B         B   D         "D  cE         eE  F         F  G         G  ,I         .I  oJ         qJ  K          \  ]          ^  ^          :_  `          ``  >a          a  db         :  y         }                        B         F             Ț         ̚             N         R             ԟ         ؟             Z         ^                        #         '  f         j                        /         3  r         v                        ;         ?  ~                    Ų             G         K             ͷ         ѷ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               4!         @!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -G-    G-G;-x    G;-I-|    I-/}-\    /}-E$-    E$--h    -]b-    ]b-낳-	    낳--h
    -K׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               I!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             8         9  q         r                                  |  \            }                                               !            F            k  C            h                                             )          *   b!         c!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            B         C                                   N         O                        
         
  Z         [                        #         $  f         g                        /         0  r         s                        ;         <  ~                                 G         H                          "         "  S#         T#  $         $  %         %  '         '  [(         `(  )         )  *         *  $,         ),  g-         l-  .         .  /         /  01         51  s2         x2  3         3  4         4  <6         A6  7         7  8         8  :         
:  H;         M;  <         <  =         =  ?         ?  T@         Y@  A         A  B         B  D         "D  `E         eE  F         F  G         G  )I         zk  l         l  m          n  Bo         Co  p         p  q         q  s         s  Nt         Ot  u         u  v         v  x         x  Zy         [y  z         z  {         {  #}         $}  f~         g~                        /         0  r         s                        ;         <  ~                                 G         H             ͎         Ύ             S         T             ٓ          B  $          f  H            l                      ң                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               )!         *!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -G-    G-);-x    );-I-|    I-}-\    }-e1-    e1--h    -b-    b-ǂ-	    ǂ--h
    -K׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               I!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             8         9  q         r                                                                                   >            ]  7          ~  X            y                                 a                                     D!         E!  }"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            >         C                                   J         O                        
         
  V         [                                 $  b         g                        +         0  n         s                        7         <  z                                  C         H                          "         "  O#         T#  $         $  %         %  '         '  _(         `(  )         )  *         *  (,         ),  k-         l-  .         .  /         /  41         51  w2         x2  3         3  4         4  @6         A6  7         7  8         8  	:         
:  L;         M;  <         <  =         =  ?         ?  X@         Y@  A         A  B         B  !D         "D  dE         eE  F         F  G         G  -I         zk  l         l  m          n  >o         Co  p         p  q         q  s         s  Jt         Ot  u         u  v         v  x         x  Vy         [y  z         z  {         {  }         $}  b~         g~                        +         0  n         s                        7         <  z                                  C         H             Ɏ         Ύ             O         T             Փ                       H  6          ~  l                        ر                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               9!         J!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -A-    A-M;-x    M;-I-|    I-5}-\    5}-)-    )--h    -cb-    cb-傳-	    傳--h
    -K׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               E!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             8         9  q         r                                  |  [            |                                               !            G            m  E            k                                             .          /   g!         h!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            A         C                        
           M         O                        
         
  Y         [                        "         $  e         g                        .         0  q         s                        :         <  }                                 F         H                          "         "  R#         T#  $         $  %         %  '         '  \(         `(  )         )  *         *  %,         ),  h-         l-  .         .  /         /  11         51  t2         x2  3         3  4         4  =6         A6  7         7  8         8  :         
:  I;         M;  <         <  =         =  ?         ?  U@         Y@  A         A  B         B  D         "D  aE         eE  F         F  G         G  *I         zk  l         l  m          n  Ao         Co  p         p  q         q  
s         s  Mt         Ot  u         u  v         v  x         x  Yy         [y  z         z  {         {  "}         $}  e~         g~                        .         0  q         s                        :         <  }                                 F         H             ̎         Ύ             R         T             ؓ            إ                      >             b  D            h                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               +!         .!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -M-    M-/;-x    /;-I-|    I-}-\    }-w7-    w7--h    -b-    b-ǂ-	    ǂ--h
    -K׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               M!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             8         9  q         r                                                                      "            B            b  ;            \            }                                 f                                     I!         J!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ?         C                                   K         O                        
         
  W         [                                  $  c         g                        ,         0  o         s                        8         <  {                                 D         H                          "         "  P#         T#  $         $  %         %  '         '  ^(         `(  )         )  *         *  ',         ),  j-         l-  .         .  /         /  31         51  v2         x2  3         3  4         4  ?6         A6  7         7  8         8  :         
:  K;         M;  <         <  =         =  ?         ?  W@         Y@  A         A  B         B   D         "D  cE         eE  F         F  G         G  ,I         zk  l         l  m          n  ?o         Co  p         p  q         q  s         s  Kt         Ot  u         u  v         v  x         x  Wy         [y  z         z  {         {   }         $}  c~         g~                        ,         0  o         s                        8         <  {                                 D         H             ʎ         Ύ             P         T             ֓                       V  D            z          µ                                 >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               9!         J!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-;-x    ;-J-|    J-k}-\    k}-/-    /--P    -b-    b-c-	    c-=-D
    =-E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               @!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 O	         P	  
         
                                    9         :  r         s                                                '            P  4          y  Y            z                          !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            >         C                                   J         O                        
         
  V         [                                 $  b         g                        +         0  n         s                        7         <  z                                  C         H                          "         2  3         3  	5         5  L6         Q6  7         7  8         8  :         :  X;         ];  <         <  =         =  !?         &?  d@         i@  A         A  B         B  -D         2D  pE         uE  F         F  G         G  9I         >I  |J         J  K         K  M         M  EN         JN  O         O  P         P  R         R  QS         VS  T         T  U         U  W         l  n         n  Uo         Zo  p         p  q         q   s         !s  bt         ct  u         u  v         v  *x         +x  my         ny  z         z  {         {  6}         7}  y~         z~                         B         C             Ȅ         Ʉ             N         O             ԉ         Չ             Z         [                        #          "            D  !          f  C            e                     v                        >                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               8!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 g
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    <     -,-p     ,--     -G-    G-6-    6-D-    D-x-t    x-[2-     [2-9-X    9--    -b-p	    b-o-x
    o-}-
    }--    -ֳ-\    ֳ--<                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               G!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         !            B            c  =            ^                                               	  	          I
  '         r                                   U         V           \                                   ?         @  x                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h         "  #         #  !%         #%  [&         ]&  '         '  (         (  	*         *  C+         E+  },         ,  -         -  .         .  +0         -0  e1         g1  2         2  3         3  5         5  M6         O6  7         7  8         8  9         9  5;         7;  o<         q<  =         =  >         >  @         @  WA         YA  B         B  C         C  E         E  ?F          h  h         :i  rj         tj  k         k  l         l   n         "n  Zo         \o  p         p  q         q  s         
s  Bt         Dt  |u         ~u  v         v  w         w  *y         ,y  dz         fz  {         {  |         |  ~         ~  L         N                                   4         6  n         p                                   V         X             ʌ         ̌                       ?  w         y                                   %  ]         _             ї         ӗ             E         G                                   -         /  g         i             ۢ         ݢ             O         Q             ç         ŧ             7         9  q         s                                 !  Y         [             Ͳ                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               F!         P!  1         1  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
                                           b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     ]
            0     -,-X     ,--     -/-    /-;-x    ;-J-|    J-k}-\    k}-/-    /--P    -b-    b-c-	    c-=-D
    =-E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               =!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 O	         P	  
         
                                    9         :  r         s                                                '            P  5          y  X            y                          !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            ?         C                                   K         O                        
         
  W         [                                  $  c         g                        ,         0  o         s                        8         <  {                                 D         H                          "         2  3         3  
5         5  M6         Q6  7         7  8         8  :         :  Y;         ];  <         <  =         =  "?         &?  e@         i@  A         A  B         B  .D         2D  qE         uE  F         F  G         G  :I         >I  }J         J  K         K  M         M  FN         JN  O         O  P         P  R         R  RS         VS  T         T  U         U  W         l  n         n  Vo         Zo  p         p  q         q  s         !s  at         ct  u         u  v         v  )x         +x  ly         ny  z         z  {         {  5}         7}  x~         z~                         A         C             Ǆ         Ʉ  
           M         O             Ӊ         Չ             Y         [             ߎ           "          ̦              ˨                      2            T  1         v                        =                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               8!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 g
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    <     -,-p     ,--     -M-    M-6-    6-D-    D-x-t    x-[2-     [2-9-X    9--    -b-p	    b-o-x
    o-}-
    }--    -ֳ-\    ֳ--<                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               K!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         !            B            c  <            ]                                               	  	          I
  (         r                                   U         V           \                                   ?         @  x                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        "  #         #  "%         #%  \&         ]&  '         '  (         (  
*         *  D+         E+  ~,         ,  -         -  .         .  ,0         -0  f1         g1  2         2  3         3  5         5  N6         O6  7         7  8         8  9         9  6;         7;  p<         q<  =         =  >         >  @         @  XA         YA  B         B  C         C  E         E  @F         h  9i         :i  sj         tj  k         k  l         l  !n         "n  [o         \o  p         p  q         q  	s         
s  Ct         Dt  }u         ~u  v         v  w         w  +y         ,y  ez         fz  {         {  |         |  ~         ~  M         N                                   5         6  o         p                                   W         X             ˌ         ̌             >         ?  x         y                        $          %            _  T                      ӗ  Ș                     G                                   .         /  h         i             ܢ         ݢ             P         Q             ħ         ŧ             8         9  r         s                                  !  Z         [             β                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               H!         P!  1         1  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
                                           b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     ^
            8     -,-h     ,--     -#-    #-A;-    A;-I-    I-)}-l    )}--    --P    --    -b-h	    b-i-p
    i-w-
    w--    -E׳-L                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ;!         >!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             8         9  q         r                                  |  \            }                                               !            F            k  C            h                                             )          *   b!         c!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h         %  &          K&  F'          '  (          (  )          *  +         W+  ,         ,  -         -  /          /  ^0         c0  1         1  2         2  '4         ,4  j5         o5  6         6  7         7  39         89  v:         {:  ;         ;  <         =  ?>         D>  ?         ?  @         @  B         B  KC         PC  D         D  E         E  G         G  WH         \H  I         I  J         J   L         %L  cM         hM  N         N  O         O  ,Q         1Q  oR         tR  S         S  T         T  8V         =V  {W         W  X         X  Z         Z  D[         I[  \         \  ]         ]  _         _  P`         U`  a         a  b         b  d         d  \e         ae  f         f  g         g  %i         *i  hj         mj  k         k  l         l  5n         6n  xo         yo  p         p  q         q  As         Bs  t         t  u         u  
w         w  Mx         Nx  y         y  z         z  |         |  Y}         Z}  ~         ~             "         #  e         f                        .         /  q         r                        :         ;  }         ~                        F         G             ̒         ͒             N         S                                b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               C!         V!  1         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
                                           b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     :
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -;-    ;-#;-x    #;-I-|    I-}-\    }--    --h    -b-    b-n-	    n--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               B!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             8         9  q         r                                                                                   >            ]  7          ~  X            y                                 a                                     D!         E!  }"                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ;         <  w         x                        +         ,  g         h             	         	             W         X                        
           F         H                                   6         8  r         t                        &         (  b         d                                     R!         T!  "         "  #         #  %          0  1          2  2          63  4          `4  E5          5  o6         :H  uI         vI  J         J  K         K  )M         *M  eN         fN  O         O  P         P  R         R  US         VS  T         T  U         U  	W         
W  EX         FX  Y         Y  Z         Z  [         [  5]         6]  q^         r^  _         _  `         `  %b         &b  ac         bc  d         d  e         e  g         g  Qh         Rh  i         i  j         j  l         l  Am         Bm  |n         ~n  o         o  p         p  0r         2r  ls         ns  t         t  u         u   w         "w  \x         ^x  y         y  z         z  |         |  L}         N}  ~         ~                         <         >  x         z                        ,         .  h         j                                   X         Z             Џ         ҏ             H                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               C!         V!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     5
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     -#-    #-A;-    A;-I-    I-)}-l    )}--    --P    --    -b-h	    b-i-p
    i-w-
    w--    -E׳-L                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               9!         >!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             8         9  q         r                                  |  [            |                                               !            G            m  E            k                                             .          /   g!         h!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        %  G&         K&  '         '  (         (  *         *  S+          W+  R,          ,  -          -  .           /  0          c0  ^1         1  2         2  (4         ,4  k5         o5  6         6  7         7  49         89  w:         {:  ;         ;  <         =  @>         D>  ?         ?  @         @  	B         B  LC         PC  D         D  E         E  G         G  XH         \H  I         I  J         J  !L         %L  dM         hM  N         N  O         O  -Q         1Q  pR         tR  S         S  T         T  9V         =V  |W         W  X         X  Z         Z  E[         I[  \         \  ]         ]  _         _  Q`         U`  a         a  b         b  d         d  ]e         ae  f         f  g         g  &i         *i  ij         mj  k         k  l         l  4n         6n  wo         yo  p         p  q         q  @s         Bs  t         t  u         u  	w         w  Lx         Nx  y         y  z         z  |         |  X}         Z}  ~         ~             !         #  d         f                        -         /  p         r                        9         ;  |         ~                        E         G             ˒         ͒             O         S                                b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               A!         V!  1         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
                                           b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     9
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -A-    A-#;-x    #;-I-|    I-}-\    }--    --h    -b-    b-n-	    n--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               D!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             8         9  q         r                                                                      "            B            b  ;            \            }                                 f                                     I!         J!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            :         <  v         x                        *         ,  f         h             	         	             V         X                                   G         H                                   7         8  s         t                        '         (  c         d                                     S!         T!  "         "  #         #  %          6  7          7  8          9  9          2:  ;          \;  A<         :H  tI         vI  J         J  K         K  (M         *M  dN         fN  O         O  P         P  R         R  TS         VS  T         T  U         U  W         
W  DX         FX  Y         Y  Z         Z  [         [  4]         6]  p^         r^  _         _  `         `  $b         &b  `c         bc  d         d  e         e  g         g  Ph         Rh  i         i  j         j  l         l  @m         Bm  }n         ~n  o         o  p         p  1r         2r  ms         ns  t         t  u         u  !w         "w  ]x         ^x  y         y  z         z  |         |  M}         N}  ~         ~                        =         >  y         z                        -         .  i         j                                   Y         Z             я         ҏ             I                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               G!         Z!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     6
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     -#-    #-6-    6-OE-    OE-x-l    x-{9-    {9--x    -{b-	    {b-c-
    c-q-
    q--|    -ֳ-    ֳ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               9!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         %            J  "          o  G            l                                  '  	          ^	  D
          
  {                      =         >  v         w                        4         5  m         n                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         l  b                                  &            d  Z         D%  &         I  J         J  4L         5L  wM         xM  N         N  O         O  @Q         AQ  R         R  S         S  	U         
U  LV         MV  W         W  X         X  Z         Z  X[         Y[  \         \  ]         ]  !_         "_  d`         e`  a         a  b         b  -d         .d  pe         qe  f         f  g         g  9i         :i  |j         }j  k         k  m         m  An         Fn  o         o  p         p  r         r  Qs         Rs  t         t  u         u  w         w  ]x         ^x  y         y  z         z  &|         '|  i}         j}  ~         ~             2         3  u         v                        >         ?             Ĉ         ň             J         K             Ѝ         э             V         W             ܒ         ݒ              ^         c                        +         ,  j         o                        3         8  v         {                        ?         D             ť         ʥ             K         P             Ѫ         ֪             W         \             ݯ                     %  c         h                        ,         1  o         t                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               8!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
                   
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            0     -,-X     ,--     -/-    /-6-x    6-D-|    D-x-\    x-w-    w--h    -b-    b-n-	    n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               >!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         !            B            c  =            ^                                               	  	          I
  '         r                                   U         V           \                                   ?         @  x                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            =         >  {         |                        5         6  s         t             	         	  -         .  k         l                        %         &  c         d                                   [         \                                   S         T                                   K          L   !         !  "         "  $         $  C%          X<  0=          x=  P>          >  p?          ?  @          @  A         G  H         H  0J         2J  nK         pK  L         L  M         M  (O         *O  fP         hP  Q         Q  R         R   T         "T  ^U         `U  V         V  W         W  Y         Y  VZ         XZ  [         [  \         \  ^         ^  N_         P_  `         `  a         a  c         
c  Fd         Hd  e         e  f         f   h         h  >i         @i  |j         ~j  k         k  l         n  o         o  *q         ,q  hr         jr  s         s  t         t  "v         $v  `w         bw  x         x  y         y  {         {  X|         Z|  }         }  ~         ~             P         R             ̃         ΃  
           H         J             Ĉ         ƈ             @         B  ~                                 8         :  v         x                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               :!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     5
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     -#-    #-6-    6-UE-    UE-x-l    x-{9-    {9--x    -ob-	    ob-c-
    c-q-
    q--|    -ֳ-    ֳ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               7!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         &            L  $          r  J            p                                  ,  	          c	  J
          
             	         
  B         C  {         |                        9         :  r         s                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                             \  R                     D%  &         I  J         J  3L         5L  vM         xM  N         N  O         O  ?Q         AQ  R         R  S         S  U         
U  KV         MV  W         W  X         X  Z         Z  W[         Y[  \         \  ]         ]   _         "_  c`         e`  a         a  b         b  ,d         .d  oe         qe  f         f  g         g  8i         :i  {j         }j  k         k  m         m  Bn         Fn  o         o  p         p  r         r  Ps         Rs  t         t  u         u  w         w  \x         ^x  y         y  z         z  %|         '|  h}         j}  ~         ~             1         3  t         v                        =         ?             È         ň             I         K             ύ         э             U         W             ے         ݒ              _         c                        *         ,  k         o                        4         8  w         {                        @         D             ƥ         ʥ  	           L         P             Ҫ         ֪             X         \             ޯ           !         %  d         h                        -         1  p         t                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               4!         @!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
                   
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            0     -,-X     ,--     -5-    5-6-x    6-D-|    D-x-\    x-w-    w--h    -b-    b-n-	    n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               @!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         !            B            c  <            ]                                               	  	          I
  (         r                                   U         V           \                                   ?         @  x                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            <         >  z         |                        4         6  r         t             	         	  ,         .  j         l                        $         &  b         d                                   Z         \                                   R         T                                   J          L   !         !  "         "  $         $  B%          A  B          C  C          8D  E          XE  0F          xF  PG         G  H         H  1J         2J  oK         pK  L         L  M         M  )O         *O  gP         hP  Q         Q  R         R  !T         "T  _U         `U  V         V  W         W  Y         Y  WZ         XZ  [         [  \         \  ^         ^  O_         P_  `         `  a         a  	c         
c  Gd         Hd  e         e  f         f  h         h  ?i         @i  }j         ~j  k         k  l         n  o         o  +q         ,q  ir         jr  s         s  t         t  #v         $v  aw         bw  x         x  y         y  {         {  Y|         Z|  }         }  ~         ~             Q         R             ̓         ΃             I         J             ň         ƈ             A         B                                   9         :  w         x                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               <!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     6
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-5;-x    5;-I-|    I-}-\    }-7-    7--h    -b-    b-	n-	    	n--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               8!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             8         9  q         r                                  |  \            }                                               !            F            k  C            h                                             )          *   b!         c!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            >         @  ~                                  >         @  ~                      	          
  >         @  ~                                  >         @  ~                                  >         @  ~                                  >         @  ~                                  >         @  ~             !         !  "          #  >$         @$  ~%          4=  >          b>  H?          ?  v@          @  A          A  B         K  L         L  /N         0N  oO         pO  P         P  Q         Q  /S         0S  oT         pT  U         U  V         V  /X         0X  oY         pY  Z         Z  [         [  /]         0]  o^         p^  _         _  `         `  /b         0b  oc         pc  d         d  e         e  /g         0g  oh         ph  i         i  j         j  /l         0l  om         pm  n         n  o         o  /q         0q  or           2         4  r         t                        2         4  r         t                        2         4  r         t                        2         4  r         t                        2         4  r         t                        2         4  r         t                        2         4  r         t                        2                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ;!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     +
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-;-x    ;-}J-|    }J-}-\    }--    --h    -b-    b-kx-	    kx--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               <!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 @	         A	  y
         z
                        7         8  p         q                                               Q  =            t                                              4   !          S!  +"          r"  J#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ?         @                                    ?         @                        	          
  ?         @                                    ?         @                                    ?         @                                    ?         @                                    ?         @               !         !  "          #  ?$         @$  %         %  &         &  '          (  ?)         @)  *         *  +         +  ,          -  ?.         @.  /         /  0         0  1          2  ?3         @3  4         4  5         5  6          7  ?8         @8  9         9  :         :  ;          <  ?=         @=  >         >  ?         ?  @          A  ?B         @B  C         C  D         D  E          F  ?G         @G  H         H  I         I  J          K  >L         @L  ~M         M  N         N  O          P  >Q         @Q  ~R         R  S         S  T          U  >V         @V  ~W         W  X         X  Y          Z  >[         @[  ~\         \  ]         ]  ^          _  >`         @`  ~a         a  b         b  c          d  >e         @e  ~f         f  g         g  h          i  >j         @j  ~k         k  l         l  m          n  o          o  p          ,q  "r          jr  `s          s  t            ^         `                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ?!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 k
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-;;-x    ;;-I-|    I-#}-\    #}-7-    7--h    -b-    b-	n-	    	n--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               8!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             8         9  q         r                                  |  [            |                                               !            G            m  E            k                                             .          /   g!         h!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ?         @                                    ?         @                        	          
  ?         @                                    ?         @                                    ?         @                                    ?         @                                    ?         @               !         !  "          #  ?$         @$  %          C   D          HD  .E          vE  \F          F  G          G  H         K  L         L  .N         0N  nO         pO  P         P  Q         Q  .S         0S  nT         pT  U         U  V         V  .X         0X  nY         pY  Z         Z  [         [  .]         0]  n^         p^  _         _  `         `  .b         0b  nc         pc  d         d  e         e  .g         0g  nh         ph  i         i  j         j  .l         0l  nm         pm  n         n  o         o  .q         0q  nr           3         4  s         t                        3         4  s         t                        3         4  s         t                        3         4  s         t                        3         4  s         t                        3         4  s         t                        3         4  s         t                        3                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ;!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     *
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-;-x    ;-}J-|    }J-}-\    }--    --h    -b-    b-kx-	    kx--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               :!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 ;	         <	  t
         u
                        2         3  k         l                                               L  9            p                                              1   	!          Q!  )"          q"  I#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            >         @  ~                                  >         @  ~                      	          
  >         @  ~                                  >         @  ~                                  >         @  ~                                  >         @  ~                                  >         @  ~             !         !  "          #  >$         @$  ~%         %  &         &  '          (  >)         @)  ~*         *  +         +  ,          -  >.         @.  ~/         /  0         0  1          2  >3         @3  ~4         4  5         5  6          7  >8         @8  ~9         9  :         :  ;          <  >=         @=  ~>         >  ?         ?  @          A  >B         @B  ~C         C  D         D  E          F  >G         @G  ~H         H  I         I  J          K  ?L         @L  M         M  N         N  O          P  ?Q         @Q  R         R  S         S  T          U  ?V         @V  W         W  X         X  Y          Z  ?[         @[  \         \  ]         ]  ^          _  ?`         @`  a         a  b         b  c          d  ?e         @e  f         f  g         g  h          i  ?j         @j  k         k  l         l  m          t  u          $v  w          bw  Xx          x  y          y  z            _         `                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               A!         R!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 k
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     --    -6-x    6-IE-|    IE-x-\    x--    --h    -{b-    {b-n-	    n--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               8!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         %            J  "          o  G            l                                  '  	          ^	  D
          
  {                      =         >  v         w                        4         5  m         n                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            <         >  z         |                        4         6  r         t             	         	  ,         .  j         l                        $         &  b         d                                   Z         \                                   R         T                                   J          L   !         !  "         "  $         $  B%         H  !J         "J  _K         `K  L         L  M         M  O         O  WP         XP  Q         Q  R         R  T         T  OU         PU  V         V  W         W  	Y         
Y  GZ         HZ  [         [  \         \  ^         ^  ?_         @_  }`         ~`  a         a  b         b  7d         8d  ue         ve  f         f  g         g  /i         0i  mj         nj  k         k  l         l  'n          n  |o         o  q         q  Er         Fr  s         s  t         t  u          v  =w         >w  {x         |x  y         y  z         z  5|         6|  s}         t}  ~         ~             -         .  k         l                        %         &  c         d             ߉                      [         \             ׎         ؎             S         T             ϓ         Г                        8            b  >            h                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               :!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     +
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-6-x    6-D-|    D-w-\    w-s-    s--h    -ub-    ub-kx-	    kx--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               <!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         !            B            c  =            ^                                               	  	          I
  '         r                                   U         V           \                                   ?         @  x                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                 $            g  _                     p&  '         '  (         (  8*         9*  {+         |+  ,         ,  .         .  D/         E/  0         0  1         1  3         3  P4         Q4  5         5  6         6  8         8  \9         ]9  :         :  ;         ;  %=         &=  h>         i>  ?         ?  @         @  1B         2B  tC         uC  D         D  E         E  =G         >G  H         H  I         I  K         K  IL         JL  M         M  N         N  P         P  QQ         VQ  R         R  S         S  U         U  ]V         bV  W         W  X         X  &Z         +Z  i[         n[  \         \  ]         ]  2_         7_  u`         z`  a         a  b          c  >d         Cd  e         e  f         f  h         h  Ji         Oi  j         j  k         k  m         m  Vn         [n  o         o  p         p  r         $r  bs         gs  t         t  u         u  +w         0w  nx         sx  y         y  z         z  7|         <|  z}         }  ~         ~              C         H             Ƀ         ΃             O         T             Ո         ڈ             [         `                        $         )  g         l                        0         5  s         x                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               :!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 k
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     --    -6-x    6-OE-|    OE-x-\    x--    --h    -ub-    ub-n-	    n--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               8!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         &            L  $          r  J            p                                  ,  	          c	  J
          
             	         
  B         C  {         |                        9         :  r         s                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            =         >  {         |                        5         6  s         t             	         	  -         .  k         l                        %         &  c         d                                   [         \                                   S         T                                   K          L   !         !  "         "  $         $  C%         H   J         "J  ^K         `K  L         L  M         M  O         O  VP         XP  Q         Q  R         R  T         T  NU         PU  V         V  W         W  Y         
Y  FZ         HZ  [         [  \         \   ^         ^  >_         @_  |`         ~`  a         a  b         b  6d         8d  te         ve  f         f  g         g  .i         0i  lj         nj  k         k  l         l  &n         o  q         q  Dr         Fr  s         s  t         t  u          v  <w         >w  zx         |x  y         y  z         z  4|         6|  r}         t}  ~         ~             ,         .  j         l                        $         &  b         d             މ                      Z         \             ֎         ؎             R         T             Γ         Г                                    
            4            ^  :                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               8!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     *
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-6-x    6-D-|    D-w-\    w-s-    s--h    -ub-    ub-kx-	    kx--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               :!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         !            B            c  <            ]                                               	  	          I
  (         r                                   U         V           \                                   ?         @  x                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                     0  (          s  k                                 p&  '         '  (         (  7*         9*  z+         |+  ,         ,   .         .  C/         E/  0         0  1         1  3         3  O4         Q4  5         5  6         6  8         8  [9         ]9  :         :  ;         ;  $=         &=  g>         i>  ?         ?  @         @  0B         2B  sC         uC  D         D  E         E  <G         >G  H         H  I         I  K         K  HL         JL  M         M  N         N  P         P  RQ         VQ  R         R  S         S  U         U  ^V         bV  W         W  X         X  'Z         +Z  j[         n[  \         \  ]         ]  3_         7_  v`         z`  a         a  b          c  ?d         Cd  e         e  f         f  h         h  Ki         Oi  j         j  k         k  m         m  Wn         [n  o         o  p         p   r         $r  cs         gs  t         t  u         u  ,w         0w  ox         sx  y         y  z         z  8|         <|  {}         }  ~         ~             D         H             ʃ         ΃             P         T             ֈ         ڈ             \         `                        %         )  h         l                        1         5  t         x                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               :!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 k
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-q;-x    q;-J-|    J-Y}-\    Y}-A-    A--h    -ib-    ib-Sx-	    Sx--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               <!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 O	         P	  
         
                                    9         :  r         s                                                '            P  4          y  Y            z                          !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            :         <  v         x                        *         ,  f         h             	         	             V         X                        
           F         H                                   6         8  r         t                        &         (  b         d                                     R!         T!  "         "  #         #  %         %  B&         D&  ~'         '  (         (  )         )  2+         4+  n,         p,  -         -  .         .  "0         $0  ^1         `1  2         2  3         3  5         5  N6         P6  7         7  8         8  :         :  >;         @;  z<         |<  =         =  >         >  .@         0@  jA         lA  B         B  C         C  E          E  ZF         \F  G         G  H         H  J         J  WK         XK  L         L  M         M  O         O  GP         HP  Q         Q  R         R  S         S  7U         8U  sV         tV  W         W  X         X  'Z         (Z  c[         d[  \         \  ]         ]  _         _  S`         T`  a         a  b         b  d         d  Ce         De  f         f  g         g  h         h  3j         4j  ok         pk  l         l  m         m  #o          ~                        H  F                      ΃  ̄                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               =!         >!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 ]
  }      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-s-     s--    -;-x    ;-wJ-|    wJ-}-\    }-7-    7--h    -b-    b--	    --h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               4!         @!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 @	         A	  y
         z
                        7         8  p         q                                               Q  =            t                                              4   !          S!  +"          r"  J#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            B         C                                   N         O                        
         
  Z         [                        #         $  f         g                        /         0  r         s                        ;         <  ~                                 G         H                          "         "  S#         T#  $         $  %         %  '         '  _(         `(  )         )  *         *  (,         ),  k-         l-  .         .  /         /  41         51  w2         x2  3         3  4         4  @6         A6  7         7  8         8  	:         
:  L;         M;  <         <  =         =  ?         ?  X@         Y@  A         A  B         B  !D         "D  dE         eE  F         F  G         G  -I         .I  pJ         qJ  K          Zc  2d          zd  Re          e  rf          f  g          g  h         n  o         o  !q         &q  dr         	  G         L             ͙         ҙ             S         X             ٞ         ޞ           !  _         d                        (         -  k         p                        4         9  w         |                        @         E             Ʊ         ˱  	           L         Q             Ҷ         ׶                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               K!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-q;-x    q;-J-|    J-Y}-\    Y}-
-    
--h    -ub-    ub-Sx-	    Sx--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               :!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 O	         P	  
         
                                    9         :  r         s                                                '            P  5          y  X            y                          !          !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ;         <  w         x                        +         ,  g         h             	         	             W         X                                   G         H                                   7         8  s         t                        '         (  c         d                                     S!         T!  "         "  #         #  %         %  C&         D&  '         '  (         (  )         )  3+         4+  o,         p,  -         -  .         .  #0         $0  _1         `1  2         2  3         3  5         5  O6         P6  7         7  8         8  :         :  ?;         @;  {<         |<  =         =  >         >  /@         0@  kA         lA  B         B  C         C  E          E  [F         \F  G         G  H         H  J         J  VK         XK  L         L  M         M  
O         O  FP         HP  Q         Q  R         R  S         S  6U         8U  rV         tV  W         W  X         X  &Z         (Z  b[         d[  \         \  ]         ]  _         _  R`         T`  a         a  b         b  d         d  Be         De  ~f         f  g         g  h         h  2j         4j  nk         pk  l         l  m         m  "o                      T  R                      ڈ  ؉                               >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               A!         B!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 ]
  }      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-y-     y--    -;-x    ;-wJ-|    wJ-}-\    }-7-    7--h    -b-    b--	    --h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               6!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 ;	         <	  t
         u
                        2         3  k         l                                               L  9            p                                              1   	!          Q!  )"          q"  I#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            A         C                        
           M         O                        
         
  Y         [                        "         $  e         g                        .         0  q         s                        :         <  }                                 F         H                          "         "  R#         T#  $         $  %         %  '         '  ^(         `(  )         )  *         *  ',         ),  j-         l-  .         .  /         /  31         51  v2         x2  3         3  4         4  ?6         A6  7         7  8         8  :         
:  K;         M;  <         <  =         =  ?         ?  W@         Y@  A         A  B         B   D         "D  cE         eE  F         F  G         G  ,I         .I  oJ         qJ  K          h  i          j  j          :k  l          Zl  2m          zm  Rn         n  o         o  "q         &q  er         	  H         L             Ι         ҙ             T         X             ڞ         ޞ           !  `         d                        )         -  l         p                        5         9  x         |                        A         E             Ǳ         ˱  
           M         Q             Ӷ         ׶                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               G!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-I-     I--    -6-x    6-E-|    E-mx-\    mx--    --h    -b-    b-Mx-	    Mx--h
    -%-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     P                       X!  1          2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     >            ]  5          |  T                                              :	  $
          o
  Y                                 T         U                                   I         J                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                     D  0          x  d                                 &  '         '  )         ")  `*         e*  +         +  ,         ,  ).         ..  l/         q/  0         0  1         1  53         :3  x4         }4  5         5  6         7  A8         F8  9         9  :         :  
<         <  M=         R=  >         >  ?         ?  A         A  YB         ^B  C         C  D         D  "F         'F  eG         jG  H         H  I         I  .K         3K  qL         vL  M         M  N         N  >P         ?P  Q         Q  R         R  T         T  JU         KU  V         V  W         W  Y         Y  VZ         WZ  [         [  \         \  ^          ^  b_         c_  `         `  a         a  +c         ,c  nd         od  e         e  f         f  7h         8h  zi         {i  j         j   l         l  Cm         Dm  n         n  o         o  q         q  Or         Pr  s         s  t         t  v         v  [w         \w  x         x  y         y  ${         %{  g|         h|  }         }  ~         ~  0         1  s         t                        <         =                      Ç             H         I             Ό         ό             T         U             ڑ         ۑ             `         a                        )                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               S!         T!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 _
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -u,-X     u,-+-     +--    -%6-x    %6-D-|    D-x-\    x--    --h    -Qb-    Qb-Yx-	    Yx--h
    -%-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     L                       D!  1          1  b=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     >            ]  ;          |  Z                                              	  
          G
  -         r                        "         #  ]         ^           p                                  !  [         \                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         D  9          ~  s                                  ,  !         &  '         '  !)         ")  d*         e*  +         +  ,         ,  -.         ..  p/         q/  0         0  1         1  93         :3  |4         }4  5         5  7         7  E8         F8  9         9  :         :  <         <  Q=         R=  >         >  ?         ?  A         A  ]B         ^B  C         C  D         D  &F         'F  iG         jG  H         H  I         I  2K         3K  uL         vL  M         M  N         N  :P         ?P  }Q         Q  R         R  T         T  FU         KU  V         V  W         W  Y         Y  RZ         WZ  [         [  \         \  ^          ^  ^_         c_  `         `  a         a  'c         ,c  jd         od  e         e  f         f  3h         8h  vi         {i  j         j  k         l  ?m         Dm  n         n  o         o  q         q  Kr         Pr  s         s  t         t  v         v  Ww         \w  x         x  y         y   {         %{  c|         h|  }         }  ~         ~  ,         1  o         t                        8         =  {                    Ç             D         I             ʌ         ό             P         U             ֑         ۑ             \         a                        %                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               3!         4!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 i
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,-C-     C--    -6-x    6-%E-|    %E-sx-\    sx--    --h    -b-    b-Mx-	    Mx--h
    -%-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     O                       T!  1          1  }=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                      @            `  8            X                                  
            ?	  *
          t
  _                                 Y         Z                                   N         O                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                      H  4           |   h!          !  "          "  #         &  '         '  )         ")  a*         e*  +         +  ,         ,  *.         ..  m/         q/  0         0  1         1  63         :3  y4         }4  5         5  6         7  B8         F8  9         9  :         :  <         <  N=         R=  >         >  ?         ?  A         A  ZB         ^B  C         C  D         D  #F         'F  fG         jG  H         H  I         I  /K         3K  rL         vL  M         M  N         N  =P         ?P  Q         Q  R         R  T         T  IU         KU  V         V  W         W  Y         Y  UZ         WZ  [         [  \         \  ^          ^  a_         c_  `         `  a         a  *c         ,c  md         od  e         e  f         f  6h         8h  yi         {i  j         j  k         l  Bm         Dm  n         n  o         o  q         q  Nr         Pr  s         s  t         t  v         v  Zw         \w  x         x  y         y  #{         %{  f|         h|  }         }  ~         ~  /         1  r         t                        ;         =  ~                    Ç             G         I             ͌         ό             S         U             ّ         ۑ             _         a                        (                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               O!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 _
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -i,-X     i,--     --    -%6-x    %6-D-|    D-x-\    x--    --h    -Wb-    Wb-Yx-	    Yx--h
    -%-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     K                       @!  1          1  [=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     >            ]  :          |  Y                                              	  
          G
  .         r                        "         #  ]         ^           p                                  !  [         \                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         f  [                                    	          N  C         &  '         '   )         ")  c*         e*  +         +  ,         ,  ,.         ..  o/         q/  0         0  1         1  83         :3  {4         }4  5         5  7         7  D8         F8  9         9  :         :  <         <  P=         R=  >         >  ?         ?  A         A  \B         ^B  C         C  D         D  %F         'F  hG         jG  H         H  I         I  1K         3K  tL         vL  M         M  N         N  ;P         ?P  ~Q         Q  R         R  T         T  GU         KU  V         V  W         W  Y         Y  SZ         WZ  [         [  \         \  ^          ^  __         c_  `         `  a         a  (c         ,c  kd         od  e         e  f         f  4h         8h  wi         {i  j         j  k         l  @m         Dm  n         n  o         o  	q         q  Lr         Pr  s         s  t         t  v         v  Xw         \w  x         x  y         y  !{         %{  d|         h|  }         }  ~         ~  -         1  p         t                        9         =  |                    Ç             E         I             ˌ         ό             Q         U             ב         ۑ             ]         a                        &                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               7!         8!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 i
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,-=-     =--    -;-x    ;-#J-|    #J-q}-\    q}-9-    9--h    -ub-    ub--	    --h
    -+-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     L                       P!  1          1  t=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 K	         L	  
         
                                   I         J                                                           A            l  H            o                                    !  !          #"  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            >         C                                   J         O                        
         
  V         [                                 $  b         g                        +         0  n         s                        7         <  z                                  C         H                          "         "  O#         T#  $         $  %         %  '         '  [(         `(  )         )  *         *  $,         ),  g-         l-  .         .  /         /  01         51  s2         x2  3         3  4         4  <6         A6  7         7  8         8  :         
:  H;         M;  <         <  =         =  ?         ?  T@         Y@  A         A  B         B  D         "D  `E         eE  F         F  G         G  )I         .I  lJ         qJ  K          L  M          M  N          O  O          4P  Q          ZQ  4R         Z             ߖ           "         #  e         f                        .         /  q         r                        :         ;  }         ~                        F         G             ̩         ͩ             R         S             خ         ٮ             ^         _                        '         (  j         k                        3                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               B!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -c,-X     c,--     --    -;-x    ;-J-|    J-}-\    }-9-    9--h    -cb-    cb--	    --h
    -+-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     I                       8!  1          1  M=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 <	         =	  w
         x
                        1         2  l         m                                               M  -            b                                              >   !          g!  ?"          "  h#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            B         C                                   N         O                        
         
  Z         [                        #         $  f         g                        /         0  r         s                        ;         <  ~                                 G         H                          "         "  S#         T#  $         $  %         %  '         '  _(         `(  )         )  *         *  (,         ),  k-         l-  .         .  /         /  41         51  w2         x2  3         3  4         4  @6         A6  7         7  8         8  	:         
:  L;         M;  <         <  =         =  ?         ?  X@         Y@  A         A  B         B  !D         "D  dE         eE  F         F  G         G  -I         .I  pJ         qJ  K          8c  d          bd  :e          e  df          f  g          g  h         Z             ۖ                    #  a         f                        *         /  m         r                        6         ;  y         ~                        B         G             ȩ         ͩ             N         S             Ԯ         ٮ             Z         _                        #         (  f         k                        /                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               6!         @!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,-=-     =--    -;-x    ;-)J-|    )J-w}-\    w}-9-    9--h    -ub-    ub--	    --h
    -+-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     M                       P!  1          1  u=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 K	         L	  
         
                                   I         J                                                           A            l  I            o                                   !  !          '"  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ?         C                                   K         O                        
         
  W         [                                  $  c         g                        ,         0  o         s                        8         <  {                                 D         H                          "         "  P#         T#  $         $  %         %  '         '  \(         `(  )         )  *         *  %,         ),  h-         l-  .         .  /         /  11         51  t2         x2  3         3  4         4  =6         A6  7         7  8         8  :         
:  I;         M;  <         <  =         =  ?         ?  U@         Y@  A         A  B         B  D         "D  aE         eE  F         F  G         G  *I         .I  mJ         qJ  K          R  ZS          S  T          T  U          U  V          W  W         Z             ޖ           !         #  d         f                        -         /  p         r                        9         ;  |         ~                        E         G             ˩         ͩ             Q         S             ׮         ٮ             ]         _                        &         (  i         k                        2                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               B!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -],-X     ],--     --    -;-x    ;-J-|    J-}-\    }-9-    9--h    -]b-    ]b--	    --h
    -+-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     H                       6!  ~1          1  I=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 7	         8	  r
         s
                        ,         -  g         h                                               H  )          }  ^                                              ;   !          e!  ="          "  g#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            A         C                        
           M         O                        
         
  Y         [                        "         $  e         g                        .         0  q         s                        :         <  }                                 F         H                          "         "  R#         T#  $         $  %         %  '         '  ^(         `(  )         )  *         *  ',         ),  j-         l-  .         .  /         /  31         51  v2         x2  3         3  4         4  ?6         A6  7         7  8         8  :         
:  K;         M;  <         <  =         =  ?         ?  W@         Y@  A         A  B         B   D         "D  cE         eE  F         F  G         G  ,I         .I  oJ         qJ  K          
i  i          4j  k          ^k  6l          l  `m          m  n         Z             ܖ                    #  b         f                        +         /  n         r                        7         ;  z         ~                         C         G             ɩ         ͩ             O         S             ծ         ٮ             [         _                        $         (  g         k                        0                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               4!         <!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,--     -5-    5-/;-x    /;-I-|    I-}-\    }--    -k-h    k-	b-    	b-ǂ-	    ǂ--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               A!         V!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             :         ;  u         v                        &            h                                                          ?            ^  6          }  U            t                     u                        %          &   `!         a!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            B         C                                   N         O                        
         
  Z         [                        #         $  f         g                        /         0  r         s                        ;         <  ~                                 G         H                          "         "  S#         T#  $         $  %         %  '         '  [(         `(  )         )  *         *  $,         ),  g-         l-  .         .  /         /  01         51  s2         x2  3         3  4         4  <6         A6  7         7  8         8  :         
:  H;         M;  <         <  =         =  ?         ?  T@         Y@  A         A  B         B  D         "D  `E         eE  F         F  G         G  )I          U  U          +V  W          PW  )X          uX  NY          Y  sZ         l  m         m  o         o  Rp         Sp  q         q  r         r  t         t  ^u         _u  v         v  w         w  'y         (y  jz         kz  {         {  |         |  3~         4~  v         w                        ?         @             Ņ         ƅ           	  K         L             ъ         Ҋ             W         X             ݏ         ޏ            !  c           ؔ         ٔ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     E                       6!  {1          1  F=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     --    -M;-    M;-I-    I-5}-d    5}--    --p    -'b-    'b-]-
    ]-k-
    k-ð-t    ð-3׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               9!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             :         ;  u         v                        &                                              =            f  >            m                                                                               	  C          D   ~!         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                        C  <                                             T#  $         $  %         %  '         '  C(         D(  )         )  *         *  +         +  3-         4-  o.         p.  /         /  0         0  #2         $2  _3         `3  4         4  5         5  7         7  O8         P8  9         9  :         :  <         <  ?=         @=  {>         |>  ?         ?  @         @  /B         0B  kC         lC  D         D  E         E  G          G  [H         \H  I         J  K         K  8M         :M  tN         vN  O         O  P         P  (R         *R  dS         fS  T         T  U         U  W         W  TX         VX  Y         Y  Z         Z  \         
\  D]         F]  ^         ^  _         _  `         `  4b         6b  pc         rc  d         d  e         e  $g         &g  `h         bh  i         i  j         j  l         l  Pm         Rm  n         n  o         o  q         q  @r         Br  |s         ~s  t         t  u         u  0w         2w  lx         nx  y         y  z         z   |         "|  \}         ^}  ~         ~                        L         N             Ą         Ƅ              <         >  x         z                        ,         .  h         j                                   X         Z                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     I                       J!  1          1  h=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
                   
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                      
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-;;-x    ;;-I-|    I-#}-\    #}--    -q-h    q-b-    b-ǂ-	    ǂ--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ?!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             :         ;  u         v                        &            h                                                           D            d  <            \            |                                             /          0   j!         k!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            A         C                        
           M         O                        
         
  Y         [                        "         $  e         g                        .         0  q         s                        :         <  }                                 F         H                          "         "  R#         T#  $         $  %         %  '         '  \(         `(  )         )  *         *  %,         ),  h-         l-  .         .  /         /  11         51  t2         x2  3         3  4         4  =6         A6  7         7  8         8  :         
:  I;         M;  <         <  =         =  ?         ?  U@         Y@  A         A  B         B  D         "D  aE         eE  F         F  G         G  *I          Z  [          [  \          	]  ]          .^  _          S_  ,`         l  m         m  o         o  Qp         Sp  q         q  r         r  t         t  ]u         _u  v         v  w         w  &y         (y  iz         kz  {         {  |         |  2~         4~  u         w                        >         @             ą         ƅ           	  J         L             Њ         Ҋ             V         X             ܏         ޏ           !  b           ה         ٔ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     F                       :!  1          1  M=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -#-    #-S;-    S;-I-    I-;}-d    ;}--    --p    -'b-    'b-]-
    ]-k-
    k-ð-t    ð-3׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ;!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             :         ;  u         v                        &                                              @            j  B            q                                                                                 H          I   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         O  H                        	          
            [  T         T#  $         $  %         %  '         '  B(         D(  ~)         )  *         *  +         +  2-         4-  n.         p.  /         /  0         0  "2         $2  ^3         `3  4         4  5         5  7         7  N8         P8  9         9  :         :  <         <  >=         @=  z>         |>  ?         ?  @         @  .B         0B  jC         lC  D         D  E         E  G          G  ZH         \H  I         J  K         K  9M         :M  uN         vN  O         O  P         P  )R         *R  eS         fS  T         T  U         U  W         W  UX         VX  Y         Y  Z         Z  	\         
\  E]         F]  ^         ^  _         _  `         `  5b         6b  qc         rc  d         d  e         e  %g         &g  ah         bh  i         i  j         j  l         l  Qm         Rm  n         n  o         o  q         q  Ar         Br  }s         ~s  t         t  u         u  1w         2w  mx         nx  y         y  z         z  !|         "|  ]}         ^}  ~         ~                        M         N             ń         Ƅ             =         >  y         z                        -         .  i         j                                   Y         Z                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     H                       J!  1          1  g=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
                   
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                      
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-;-x    ;-/J-|    /J-}}-\    }}-+-    +--P    -?b-    ?b-c-	    c-+-D
    +-9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               >!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 K	         L	  
         
                                   I         J                                                           A            l  H            o                                    !  !          #"  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            >         C                                   J         O                        
         
  V         [                                 $  b         g                        +         0  n         s                        7         <  z                                  C         H                          "         "  O#         T#  $         $  %         %  '         '  [(         `(  )         )  *         *  $,         ),  g-         l-  .         .  /         /  01         51  s2         x2  3         3  4         4  <6         A6  7         7  8         8  :         
:  H;         M;  <         <  =         =  ?         ?  T@         Y@  A         A  B         B  D         ng  h         h  i         i  2k         7k  ul         zl  m         m  n          o  >p         Cp  q         q  r         r  t         t  Ju         Ou  v         v  w         w  y         y  Sz         Xz  {         {  |         |  ~         ~  ]         b                        &         +  l         m                        2         3  t         u                        =         >           4  r         w              }          ħ              ݩ          $            T  =                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     K                       T!  1          1  y=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 g
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    <     -,-p     ,--     -#-    #-16-    16-D-    D-x-t    x-G4-     G4-;-X    ;--    -{b-p	    {b-c-x
    c-q-
    q--    -ֳ-\    ֳ--<                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ;!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     >            ]  ;          |  Z                                              	  
          G
  -         r                        "         #  ]         ^           p                                  !  [         \                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        (#  c$         CF  ~G         G  H         H  I         I  8K         :K  vL         xL  M         M  N         N  0P         2P  nQ         pQ  R         R  S         S  (U         *U  fV         hV  W         W  X         X   Z         "Z  ^[         `[  \         \  ]         ]  _         _  V`         X`  a         a  b         b  d         d  Ne         Pe  f         f  g         g  i         
i  Fj         Hj  k         k  l         l   n         n  >o         @o  |p         ~p  q         q  r         r  6t         8t  tu         vu  v         v  w         w  .y         0y  lz         nz  {         {  |         |  &~         (~  d         f                                    \         ^             ؆         چ             T         V             Ћ         ҋ             L         N             Ȑ         ʐ             C         E                                   9         ;  w         y                                   3  *          q  h                                 +  g         i                        !         #  _         a             ۪         ݪ             W         Y             ӯ         կ             O         Q             ˴                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               <!         L!  1         1  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
                                           b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            0     -,-X     ,--     -)-    )-;-x    ;-/J-|    /J-}}-\    }}-1-    1--P    -?b-    ?b-c-	    c-1-D
    1-9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               =!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 K	         L	  
         
                                   I         J                                                           A            l  I            o                                   !  !          '"  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            ?         C                                   K         O                        
         
  W         [                                  $  c         g                        ,         0  o         s                        8         <  {                                 D         H                          "         "  P#         T#  $         $  %         %  '         '  \(         `(  )         )  *         *  %,         ),  h-         l-  .         .  /         /  11         51  t2         x2  3         3  4         4  =6         A6  7         7  8         8  :         
:  I;         M;  <         <  =         =  ?         ?  U@         Y@  A         A  B         B  D         ng  h         h  i         i  3k         7k  vl         zl  m         m  n          o  ?p         Cp  q         q  r         r  t         t  Ku         Ou  v         v  w         w  y         y  Tz         Xz  {         {  |         |  ~         ~  ^         b                        '         +  k         m                        1         3  s         u                        <         >           4  s         w              m                        ͯ                      D  -                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     L                       T!  1          1  z=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 h
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    <     -,-p     ,--     -)-    )-16-    16-D-    D-x-t    x-G4-     G4-;-X    ;--    -{b-p	    {b-c-x
    c-q-
    q--    -ֳ-\    ֳ--<                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               =!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     >            ]  :          |  Y                                              	  
          G
  .         r                        "         #  ]         ^           p                                  !  [         \                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        (#  d$         CF  G         G  H         H  I         I  9K         :K  wL         xL  M         M  N         N  1P         2P  oQ         pQ  R         R  S         S  )U         *U  gV         hV  W         W  X         X  !Z         "Z  _[         `[  \         \  ]         ]  _         _  W`         X`  a         a  b         b  d         d  Oe         Pe  f         f  g         g  	i         
i  Gj         Hj  k         k  l         l  n         n  ?o         @o  }p         ~p  q         q  r         r  7t         8t  uu         vu  v         v  w         w  /y         0y  mz         nz  {         {  |         |  '~         (~  e         f                                    ]         ^             ن         چ             U         V             ы         ҋ             M         N             ɐ         ʐ             D         E                                   :         ;  x         y                        2         3  p         q                        *          +  "          i  `                        ܦ          #           a             ܪ         ݪ             X         Y             ԯ         կ             P         Q             ̴                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               <!         L!  1         1  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
                                           b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            4     -,-`     ,--     -)-    )-5;-    5;-I-    I-}-d    }-%-    %--p    -b-    b-i-
    i-w-
    w-۰-t    ۰-?׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ;!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             :         ;  u         v                        &            h                                                          ?            ^  6          }  U            t                     u                        %          &   `!         a!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                        @  6            v                                  d#  $         $  %         %  "'         $'  b(         d(  )         )  *         *  ",         $,  b-         d-  .         .  /         /  "1         $1  b2         d2  3         3  4         4  "6         $6  b7         d7  8         8  9         9  ";         $;  b<         d<  =         =  >         >  "@         $@  bA         dA  B         B  C         C  "E         $E  bF         dF  G         G  H         H  "J         $J  cK         dK  L         L  M         M  #O         $O  cP         dP  Q         Q  R         R  #T         $T  cU         dU  V         V  W         W  #Y         $Y  cZ         dZ  [         [  \         \  #^         $^  c_         d_  `         `  a         a  #c         $c  cd         dd  e         e  f         f  #h         $h  ci         di  j         j  k         k  #m         $m  cn         dn  o         ds  t         t  u         u  "w         $w  bx         dx  y         y  z         z  "|         $|  b}         d}  ~         ~             "         $  b         d                        "         $  b         d                        "         $  b         d                        "         $  b         d                        "         $  b         d                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               G!         b!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
                   
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     .
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-S;-x    S;-I-|    I-;}-\    ;}-2-    2--h    -b-    b-n-	    n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               <!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             :         ;  u         v                        &                                              =            f  >            m                                                                               	  C          D   ~!         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ?         @                                    ?         @                        	          
  ?         @                                    ?         @                                    ?         @                                    ?         @                                    ?         @               !          :  ;          *<  =          X=  ?>          >  m?          ?  @         F  H         H  GI         I  J         J  L         L  QM         SM  N         N  O         O  Q         Q  QR         SR  S         S  T         T  V         V  QW         SW  X         X  Y         Y  [         [  Q\         S\  ]         ]  ^         ^  `         `  Qa         Sa  b         b  c         c  e         e  Qf         Sf  g         g  h         h  j         j  Qk         Sk  l         l  m         m  o         o  p         p  #r         $r  cs           א         ؐ             W         X             ו         ؕ             W         X             ך         ؚ             W         X             ן         ؟             W         X             פ         ؤ             W         X             ש         ة             W         X             ׮         خ             W         X                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ?!         R!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     1
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -)-    )-A;-    A;-I-    I-)}-d    )}-%-    %--p    -b-    b-i-
    i-w-
    w-հ-t    հ-?׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ;!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             :         ;  u         v                        &            h                                                           D            d  <            \            |                                             /          0   j!         k!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         @  6            v            	           
  
          @  6         d#  $         $  %         %  #'         $'  c(         d(  )         )  *         *  #,         $,  c-         d-  .         .  /         /  #1         $1  c2         d2  3         3  4         4  #6         $6  c7         d7  8         8  9         9  #;         $;  c<         d<  =         =  >         >  #@         $@  cA         dA  B         B  C         C  #E         $E  cF         dF  G         G  H         H  #J         $J  bK         dK  L         L  M         M  "O         $O  bP         dP  Q         Q  R         R  "T         $T  bU         dU  V         V  W         W  "Y         $Y  bZ         dZ  [         [  \         \  "^         $^  b_         d_  `         `  a         a  "c         $c  bd         dd  e         e  f         f  "h         $h  bi         di  j         j  k         k  "m         $m  bn         dn  o         ds  t         t  u         u  #w         $w  cx         dx  y         y  z         z  #|         $|  c}         d}  ~         ~             #         $  c         d                        #         $  c         d                        #         $  c         d                        #         $  c         d                        #         $  c         d                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               G!         b!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
                   
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     -
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-Y;-x    Y;-I-|    I-A}-\    A}-2-    2--h    -b-    b-n-	    n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               <!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             :         ;  u         v                        &                                              @            j  B            q                                                                                 H          I   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            >         @  ~                                  >         @  ~                      	          
  >         @  ~                                  >         @  ~                                  >         @  ~                                  >         @  ~                                  >         @  ~             !          @  A          B  B          >C  %D          lD  SE          E  F         F  H         H  FI         I  J         J  L         L  RM         SM  N         N  O         O  Q         Q  RR         SR  S         S  T         T  V         V  RW         SW  X         X  Y         Y  [         [  R\         S\  ]         ]  ^         ^  `         `  Ra         Sa  b         b  c         c  e         e  Rf         Sf  g         g  h         h  j         j  Rk         Sk  l         l  m         m  o         o  p         p  "r         $r  bs           ֐         ؐ             V         X             ֕         ؕ             V         X             ֚         ؚ             V         X             ֟         ؟             V         X             ֤         ؤ             V         X             ֩         ة             V         X             ֮         خ             V         X                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               A!         V!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     2
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     -5-    5-6-    6-7E-    7E-x-l    x-8-    8--x    -b-	    b-i-
    i-w-
    w--|    -ֳ-    ֳ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               =!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     >            ]  5          |  T                                              :	  $
          o
  Y                                 T         U                                   I         J                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         1  2          2  3          4   5          J5  06          z6  `7         zI  J         J  K         K  2M         4M  pN         rN  O         O  P         P  *R         ,R  hS         jS  T         T  U         U  "W         $W  `X         bX  Y         Y  Z         Z  \         \  X]         Z]  ^         ^  _         _  a         a  Pb         Rb  c         c  d         d  
f         f  Hg         Jg  h         h  i         i  k         k  Al         Bl  m         m  n         n  o         o  9q         :q  wr         xr  s         s  t         t  1v         2v  ow         pw  x         x  y         y  ){         *{  g|         h|  }         }  ~         ~  !         "  _         `             ۃ         ܃             W         X             ӈ         Ԉ             O         P             ˍ         ̍  	         
  G         H             Ò         Ē             ?         @  }         ~                        7         8  u         v                        /         0  m         n                        '         (  e         f                                    ]         ^             ٬         ڬ             U         V             ѱ         ұ             L         N             ȶ         ޶             Y                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               >!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
                   
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     j
            0     -,-X     ,--     -)-    )-76-x    76-D-|    D-x-\    x--    --h    -b-    b-n-	    n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               :!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     >            ]  ;          |  Z                                              	  
          G
  -         r                        "         #  ]         ^           p                                  !  [         \                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            B         C                                   N         O                        
         
  Z         [                        #         $  f         g                        /         0  r         s                        ;         <  ~                                 G         H                          "         "  S#         T#  $         $  %         %  '         '  [(         `(  )         )  *         *  $,         ),  g-         l-  .         .  /         /  01         51  s2         x2  3         3  4         4  <6         A6  7         7  8         8  :         
:  H;         M;  <         <  =         =  ?         ?  T@         Y@  A         A  B         B  D         "D  `E         eE  F          F  G          G  H          I  J          HJ  0K          rK  ZL         i  j         j  7l         <l  zm         m  n         n   p         p  Cq         Hq  r         r  s         s  u         u  Ov         Tv  w         w  x         x  z         z  [{         `{  |         |  }         }  $         )  g         l                        0         5  s         x                        <         A                      ǋ           
  H         M             ΐ         Ӑ             T         Y                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               :!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     1
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     -5-    5-6-    6-7E-    7E-x-l    x-8-    8--x    -b-	    b-i-
    i-w-
    w--|    -ֳ-    ֳ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               =!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                      @            `  8            X                                  
            ?	  *
          t
  _                                 Y         Z                                   N         O                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         7  8          8  9          
:  :          :;   <          j<  P=         zI  J         J  K         K  3M         4M  qN         rN  O         O  P         P  +R         ,R  iS         jS  T         T  U         U  #W         $W  aX         bX  Y         Y  Z         Z  \         \  Y]         Z]  ^         ^  _         _  a         a  Qb         Rb  c         c  d         d  f         f  Ig         Jg  h         h  i         i  k         k  @l         Bl  ~m         m  n         n  o         o  8q         :q  vr         xr  s         s  t         t  0v         2v  nw         pw  x         x  y         y  ({         *{  f|         h|  }         }  ~         ~            "  ^         `             ڃ         ܃             V         X             ҈         Ԉ             N         P             ʍ         ̍           
  F         H                      Ē              >         @  |         ~                        6         8  t         v                        .         0  l         n                        &         (  d         f                                    \         ^             ج         ڬ             T         V             б         ұ             M         N             ɶ         ޶             X                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               <!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
                   
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     i
            0     -,-X     ,--     -)-    )-76-x    76-D-|    D-x-\    x--    --h    -b-    b-n-	    n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               :!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     >            ]  :          |  Y                                              	  
          G
  .         r                        "         #  ]         ^           p                                  !  [         \                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            A         C                        
           M         O                        
         
  Y         [                        "         $  e         g                        .         0  q         s                        :         <  }                                 F         H                          "         "  R#         T#  $         $  %         %  '         '  \(         `(  )         )  *         *  %,         ),  h-         l-  .         .  /         /  11         51  t2         x2  3         3  4         4  =6         A6  7         7  8         8  :         
:  I;         M;  <         <  =         =  ?         ?  U@         Y@  A         A  B         B  D         "D  aE         eE  F          L  M          M  N          N  O          P  Q          DQ  ,R         i  j         j  8l         <l  {m         m  n         n  p         p  Dq         Hq  r         r  s         s  u         u  Pv         Tv  w         w  x         x  z         z  \{         `{  |         |  }         }  %         )  h         l                        1         5  t         x                        =         A             Ë         ǋ           
  I         M             ϐ         Ӑ             U         Y                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               :!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     2
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -A-    A-A;-x    A;-I-|    I-)}-\    )}--9-    -9--h    -b-    b-n-	    n--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               B!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             :         ;  u         v                        &            h                                                          ?            ^  6          }  U            t                     u                        %          &   `!         a!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            :         <  v         x                        *         ,  f         h             	         	             V         X                        
           F         H                                   6         8  r         t                        &         (  b         d                                     R!         T!  "         "  #         #  %         %  C&         D&  '         '  (         (  )         )  3+         4+  o,         p,  -         -  .         .  #0         $0  _1         `1  2         2  3         3  5         5  O6         P6  7         7  8         8  :         :  ?;         @;  {<         |<  =         =  >         >  /@         0@  kA         lA  B         B  C         C  E          E  [F         \F  G          k^  C_          _  e`          `  a          a  b          b  c         m  o         o  Op           ֕         ؕ             N         P             ƚ         Ț             >         @  z         |                        .         0  j         l                                    Z         \             Ҫ         Ԫ             J         L             ¯         į              :         <  v         x                        *         ,  f         h                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               E!         Z!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     +
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-<-x    <-J-|    J-}-\    }-c6-    c6--h    -b-    b-qx-	    qx--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               >!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 <	         =	  w
         x
                        1         2  l         m                                               M  -            b                                              >   !          g!  ?"          "  h#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            B         C                                   N         O                        
         
  Z         [                        #         $  f         g                        /         0  r         s                        ;         <  ~                                 G         H                          "         "  S#         T#  $         $  %          x2  f3          3  4          4  5          A6  /7          7  r8         F  G         G  -I         .I  pJ         qJ  K         K  L         L  9N         :N  |O         }O  P         P  R         R  ES         FS  T         T  U         U  W         W  QX         RX  Y         Y  Z         Z  \         \  ]]         ^]  ^         ^  _         _  &a         'a  ib         jb  c         c  d         d  2f         3f  ug         vg  h         h  i         i  >k         ?k  l         l  m         m  o         o  Fp         Kp  q         q  r         r  t         t  Ru         Wu  v         v  w         w  y          y  ^z         cz  {         {  |         |  '~         ,~  j         o                        3         8  v         {                        ?         D             Ŋ         ʊ             K         ұ             S         X             ٶ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ;!         J!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 i
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -{ -X     { -ְ-     ְ-Cܰ-    Cܰ-]4-p    ]4-@-T    @-h-4    h--9-    -9--8    -b-    b-n-	    n--8
    -E׳-
                   =         b      J      7   U   7   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               D!         T!  R1                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 64 })  outputElements  5184    input0Elements  31104   input1Elements  24576   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     4                           #         b      J      7   M     operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       @             :         ;  u         v                       @	  
          d
  <            `                                                          0                        A         B  |         }                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     4                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  1728    input0Elements  1728    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       x            ;         <  w         x                        +         ,  g         h             	         	             W         X                                   G         H                                   7         8  s         t                        '         (  c         d                                     S!         T!  "         "  #         #  %         %  B&         D&  ~'         '  (         (  )         )  2+         4+  n,         p,  -         -  .         .  "0         $0  ^1         `1  2         2  3         3  5         5  N6         P6  7         7  8         8  :         :  >;         @;  z<         |<  =         =  >         >  .@         0@  jA         lA  B         B  C         C  E          E  ZF         \F  G          d  d          7e  f          Yf  1g          {g  Sh         m  o         o  Np           ו         ؕ             O         P             ǚ         Ț             ?         @  {         |                        /         0  k         l                                    [         \             Ӫ         Ԫ             K         L             ï         į              ;         <  w         x                        +         ,  g         h                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               E!         Z!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     *
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-<-x    <-J-|    J-}-\    }-c6-    c6--h    -b-    b-qx-	    qx--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               :!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 7	         8	  r
         s
                        ,         -  g         h                                               H  )          }  ^                                              ;   !          e!  ="          "  g#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            A         C                        
           M         O                        
         
  Y         [                        "         $  e         g                        .         0  q         s                        :         <  }                                 F         H                          "         "  R#         T#  $         $  %          8  9          
:  :          M;  ;<          <  ~=          =  >         F  G         G  ,I         .I  oJ         qJ  K         K  L         L  8N         :N  {O         }O  P         P  R         R  DS         FS  T         T  U         U  W         W  PX         RX  Y         Y  Z         Z  \         \  \]         ^]  ^         ^  _         _  %a         'a  hb         jb  c         c  d         d  1f         3f  tg         vg  h         h  i         i  =k         ?k  l         l  m         m  o         o  Gp         Kp  q         q  r         r  t         t  Su         Wu  v         v  w         w  y          y  _z         cz  {         {  |         |  (~         ,~  k         o                        4         8  w         {                        @         D             Ɗ         ʊ  	           L         ұ             T         X             ڶ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               =!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 i
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -G-    G-6-x    6-IE-|    IE-x-\    x--    --h    -b-    b-kx-	    kx--h
    -K׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               D!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         "            D            f  >            `                                    	          I	  8
          ~
  m                      (         )  c         d                                   X         Y                        	                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                        8  &          p  ^                                 %  2'         7'  u(         z(  )         )  *          +  >,         C,  -         -  .         .  0         0  J1         O1  2         2  3         3  5         5  V6         [6  7         7  8         8  :         $:  b;         g;  <         <  =         =  +?         0?  n@         s@  A         A  B         B  7D         <D  zE         E  F         F   H         H  CI         HI  J         J  K         K  M         M  SN         TN  O         O  P         P  R         R  _S         `S  T         T  U         U  (W         )W  kX         lX  Y         Y  Z         Z  4\         5\  w]         x]  ^         ^  _         _  @a         Aa  b         b  c         c  	e         
e  Lf         Mf  g         g  h         h  j         j  Xk         Yk  l         l  m         m  !o         "o  dp         ep  q         q  r         r  -t         .t  pu         qu  v         v  w         w  9y         :y  |z         }z  {         {  }         }  E~         F~             ˀ         ̀             Q         R             ׅ         ؅             ]         ^                        &         '  i         j                        2         3  u         v                        >         ?                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               H!         X!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 _
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -;-    ;-=6-x    =6-D-|    D-%x-\    %x--    --h    -ub-    ub-wx-	    wx--h
    -K׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               @!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     >            ]  6          |  U            |                                  	  	          G
  (         r                        "         #  ]         ^           p                                  !  [         \                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         <  5            x                                  H  A         %  6'         7'  y(         z(  )         )  *          +  B,         C,  -         -  .         .  0         0  N1         O1  2         2  3         3  5         5  Z6         [6  7         7  8         8  #:         $:  f;         g;  <         <  =         =  /?         0?  r@         s@  A         A  B         B  ;D         <D  ~E         E  F         F  H         H  GI         HI  J         J  K         K  M         M  ON         TN  O         O  P         P  R         R  [S         `S  T         T  U         U  $W         )W  gX         lX  Y         Y  Z         Z  0\         5\  s]         x]  ^         ^  _         _  <a         Aa  b         b  c         c  e         
e  Hf         Mf  g         g  h         h  j         j  Tk         Yk  l         l  m         m  o         "o  `p         ep  q         q  r         r  )t         .t  lu         qu  v         v  w         w  5y         :y  xz         }z  {         {  |         }  A~         F~             ǀ         ̀  
           M         R             Ӆ         ؅             Y         ^             ߊ           "         '  e         j                        .         3  q         v                        :         ?  }                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               4!         <!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 i
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -M-    M-6-x    6-IE-|    IE-x-\    x--    --h    -b-    b-kx-	    kx--h
    -K׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               F!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         !            B            c  ;            \                                              D	  2
          y
  g                      #         $  ^         _                                   S         T                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                     P  >            v	          	  
          
           %  3'         7'  v(         z(  )         )  *          +  ?,         C,  -         -  .         .  0         0  K1         O1  2         2  3         3  5         5  W6         [6  7         7  8         8   :         $:  c;         g;  <         <  =         =  ,?         0?  o@         s@  A         A  B         B  8D         <D  {E         E  F         F  H         H  DI         HI  J         J  K         K  M         M  RN         TN  O         O  P         P  R         R  ^S         `S  T         T  U         U  'W         )W  jX         lX  Y         Y  Z         Z  3\         5\  v]         x]  ^         ^  _         _  ?a         Aa  b         b  c         c  e         
e  Kf         Mf  g         g  h         h  j         j  Wk         Yk  l         l  m         m   o         "o  cp         ep  q         q  r         r  ,t         .t  ou         qu  v         v  w         w  8y         :y  {z         }z  {         {  }         }  D~         F~             ʀ         ̀             P         R             օ         ؅             \         ^                        %         '  h         j                        1         3  t         v                        =         ?                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               F!         T!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 _
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-=6-x    =6-D-|    D-%x-\    %x--    --h    -{b-    {b-wx-	    wx--h
    -K׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               <!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     >            ]  5          |  T            }                                  	  	          G
  )         r                        "         #  ]         ^           p                                  !  [         \                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                         !          "  
#          T#  M$          $  %         %  5'         7'  x(         z(  )         )  *          +  A,         C,  -         -  .         .  
0         0  M1         O1  2         2  3         3  5         5  Y6         [6  7         7  8         8  ":         $:  e;         g;  <         <  =         =  .?         0?  q@         s@  A         A  B         B  :D         <D  }E         E  F         F  H         H  FI         HI  J         J  K         K  M         M  PN         TN  O         O  P         P  R         R  \S         `S  T         T  U         U  %W         )W  hX         lX  Y         Y  Z         Z  1\         5\  t]         x]  ^         ^  _         _  =a         Aa  b         b  c         c  e         
e  If         Mf  g         g  h         h  j         j  Uk         Yk  l         l  m         m  o         "o  ap         ep  q         q  r         r  *t         .t  mu         qu  v         v  w         w  6y         :y  yz         }z  {         {  |         }  B~         F~             Ȁ         ̀             N         R             ԅ         ؅             Z         ^                        #         '  f         j                        /         3  r         v                        ;         ?  ~                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               6!         @!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 i
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     ---X     ---     -q-    q-;-x    ;-J-|    J-}-\    }-8-    8--h    -ub-    ub-߂-	    ߂--h
    -Q׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               [!         \!  	2         
2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 K	         L	  
         
                                   I         J                                               $            V  .            `                                                D!  "          r"  J#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            >         C                                   J         O                        
         
  V         [                                 $  b         g                        +         0  n         s                        7         <  z                                  C         H                          "         "  O#         T#  $         $  %         %  '         '  [(         `(  )         )  *         *  $,         ),  g-         l-  .         .  /         /  01         51  s2         x2  3         3  4         4  <6         A6  7         7  8         8  :         
:  H;         M;  <         <  =         =  ?         ?  T@         Y@  A         A  B         B  D         "D  `E         eE  F         F  G         G  )I         .I  lJ         qJ  K          *{  |          ^|  ;}          }  o~          ~              ׀         :  |         }                        E         F             ˚         ̚             Q         R             ן         ؟             ]         ^                        &         '  i         j                        2         3  u         v                        >         ?             Ĳ         Ų             J         K             з         ѷ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               2!         8!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-y-     y--    -=<-x    =<-J-|    J-%~-\    %~-8-    8--h    -b-    b-߂-	    ߂--h
    -Q׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               .!         0!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 <	         =	  w
         x
                        1         2  l         m                                               M  ,            a                                              P   (!          !  Z"          "  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            B         C                                   N         O                        
         
  Z         [                        #         $  f         g                        /         0  r         s                        ;         <  ~                                 G         H                          "         "  S#         T#  $         $  %         %  '         '  _(         `(  )         )  *         *  (,         ),  k-         l-  .         .  /         /  41         51  w2         x2  3         3  4         4  @6         A6  7         7  8         8  	:         
:  L;         M;  <         <  =         =  ?         ?  X@         Y@  A         A  B         B  !D         "D  dE         eE  F         F  G         G  -I         .I  pJ         qJ  K          K  L          L  M           N  N          &O  P          LP  (Q         :  x         }                        A         F             ǚ         ̚  
           M         R             ӟ         ؟             Y         ^             ߤ           "         '  e         j                        .         3  q         v                        :         ?  }                    Ų             F         K             ̷         ѷ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               :!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     ---X     ---     -e-    e-;-x    ;-J-|    J-}-\    }-8-    8--h    -{b-    {b-߂-	    ߂--h
    -Q׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               W!         X!  2         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 K	         L	  
         
                                   I         J                                               "            S  +            \                                                B!  "          q"  I#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ?         C                                   K         O                        
         
  W         [                                  $  c         g                        ,         0  o         s                        8         <  {                                 D         H                          "         "  P#         T#  $         $  %         %  '         '  \(         `(  )         )  *         *  %,         ),  h-         l-  .         .  /         /  11         51  t2         x2  3         3  4         4  =6         A6  7         7  8         8  :         
:  I;         M;  <         <  =         =  ?         ?  U@         Y@  A         A  B         B  D         "D  aE         eE  F         F  G         G  *I         .I  mJ         qJ  K          .            b  ?            s          ʄ              ۆ         :  {         }                        D         F             ʚ         ̚             P         R             ֟         ؟             \         ^                        %         '  h         j                        1         3  t         v                        =         ?             ò         Ų             I         K             Ϸ         ѷ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               4!         <!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-s-     s--    -1<-x    1<-J-|    J-~-\    ~-8-    8--h    -b-    b-߂-	    ߂--h
    -Q׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               -!         .!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 7	         8	  r
         s
                        ,         -  g         h                                               H  &          }  [                                              I   !!          z!  R"          "  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            A         C                        
           M         O                        
         
  Y         [                        "         $  e         g                        .         0  q         s                        :         <  }                                 F         H                          "         "  R#         T#  $         $  %         %  '         '  ^(         `(  )         )  *         *  ',         ),  j-         l-  .         .  /         /  31         51  v2         x2  3         3  4         4  ?6         A6  7         7  8         8  :         
:  K;         M;  <         <  =         =  ?         ?  W@         Y@  A         A  B         B   D         "D  cE         eE  F         F  G         G  ,I         .I  oJ         qJ  K          rQ  NR          R  tS          S  T          T  U          
V  V         :  y         }                        B         F             Ț         ̚             N         R             ԟ         ؟             Z         ^                        #         '  f         j                        /         3  r         v                        ;         ?  ~                    Ų             G         K             ͷ         ѷ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               8!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -M-    M-Y;-x    Y;-I-|    I-A}-\    A}-/-    /--h    -cb-    cb-߂-	    ߂--h
    -K׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               M!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             :         ;  u         v                        &            l                                                          +            R  *          y  Q            x                                             9          :   t!         u!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            B         C                                   N         O                        
         
  Z         [                        #         $  f         g                        /         0  r         s                        ;         <  ~                                 G         H                          "         "  S#         T#  $         $  %         %  '         '  [(         `(  )         )  *         *  $,         ),  g-         l-  .         .  /         /  01         51  s2         x2  3         3  4         4  <6         A6  7         7  8         8  :         
:  H;         M;  <         <  =         =  ?         ?  T@         Y@  A         A  B         B  D         "D  `E         eE  F         F  G         G  )I         zk  l         l  m          n  Bo         Co  p         p  q         q  s         s  Nt         Ot  u         u  v         v  x         x  Zy         [y  z         z  {         {  #}         $}  f~         g~                        /         0  r         s                        ;         <  ~                                 G         H             ͎         Ύ             S         T             ٓ                      Ϋ              ҭ                      :                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               -!         .!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -M-    M-G;-x    G;-I-|    I-/}-\    /}-M%-    M%--h    -b-    b-ǂ-	    ǂ--h
    -K׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               M!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             :         ;  u         v                        &                                              %            F            g  ?            `                                             u                        %          &   `!         a!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            >         C                                   J         O                        
         
  V         [                                 $  b         g                        +         0  n         s                        7         <  z                                  C         H                          "         "  O#         T#  $         $  %         %  '         '  _(         `(  )         )  *         *  (,         ),  k-         l-  .         .  /         /  41         51  w2         x2  3         3  4         4  @6         A6  7         7  8         8  	:         
:  L;         M;  <         <  =         =  ?         ?  X@         Y@  A         A  B         B  !D         "D  dE         eE  F         F  G         G  -I         zk  l         l  m          n  >o         Co  p         p  q         q  s         s  Jt         Ot  u         u  v         v  x         x  Vy         [y  z         z  {         {  }         $}  b~         g~                        +         0  n         s                        7         <  z                                  C         H             Ɏ         Ύ             O         T             Փ                      ,            b  N                      Τ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               =!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -G-    G-_;-x    _;-I-|    I-G}-\    G}-a5-    a5--h    -ib-    ib-߂-	    ߂--h
    -K׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               I!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             :         ;  u         v                        &            k                                                          +            S  +          {  S            {                                             >          ?   y!         z!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            A         C                        
           M         O                        
         
  Y         [                        "         $  e         g                        .         0  q         s                        :         <  }                                 F         H                          "         "  R#         T#  $         $  %         %  '         '  \(         `(  )         )  *         *  %,         ),  h-         l-  .         .  /         /  11         51  t2         x2  3         3  4         4  =6         A6  7         7  8         8  :         
:  I;         M;  <         <  =         =  ?         ?  U@         Y@  A         A  B         B  D         "D  aE         eE  F         F  G         G  *I         zk  l         l  m          n  Ao         Co  p         p  q         q  
s         s  Mt         Ot  u         u  v         v  x         x  Yy         [y  z         z  {         {  "}         $}  e~         g~                        .         0  q         s                        :         <  }                                 F         H             ̎         Ύ             R         T             ؓ          ^  >            b                      ʳ              ε                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               /!         2!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     ---X     ---     -Y-    Y-S;-x    S;-I-|    I-;}-\    ;}-Y+-    Y+--h    -b-    b-ǂ-	    ǂ--h
    -K׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               Q!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             :         ;  u         v                        &                                              (             J  "          l  D            f                                                                     /          0   j!         k!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ?         C                                   K         O                        
         
  W         [                                  $  c         g                        ,         0  o         s                        8         <  {                                 D         H                          "         "  P#         T#  $         $  %         %  '         '  ^(         `(  )         )  *         *  ',         ),  j-         l-  .         .  /         /  31         51  v2         x2  3         3  4         4  ?6         A6  7         7  8         8  :         
:  K;         M;  <         <  =         =  ?         ?  W@         Y@  A         A  B         B   D         "D  cE         eE  F         F  G         G  ,I         zk  l         l  m          n  ?o         Co  p         p  q         q  s         s  Kt         Ot  u         u  v         v  x         x  Wy         [y  z         z  {         {   }         $}  c~         g~                        ,         0  o         s                        8         <  {                                 D         H             ʎ         Ύ             P         T             ֓                      :  &          p  \                      ܪ  ȫ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               =!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -;-    ;-;-x    ;-/J-|    /J-}}-\    }}-/-    /--P    -b-    b-c-	    c-=-D
    =-K׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               D!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 K	         L	  
         
                                   I         J                                                           A  %          l  P            s                                      !          "  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            >         C                                   J         O                        
         
  V         [                                 $  b         g                        +         0  n         s                        7         <  z                                  C         H                          "          $  %          L%  9&          &  m'          '  (          (  )         2  3         3  	5         5  L6         Q6  7         7  8         8  :         :  X;         ];  <         <  =         =  !?         &?  d@         i@  A         A  B         B  -D         2D  pE         uE  F         F  G         G  9I         >I  |J         J  K         K  M         M  EN         JN  O         O  P         P  R         R  QS         VS  T         T  U         U  W         l  n         n  Uo         Zo  p         p  q         q   s         !s  bt         ct  u         u  v         v  *x         +x  my         ny  z         z  {         {  6}         7}  y~         z~                         B         C             Ȅ         Ʉ             N         O             ԉ         Չ             Z         [                        #         v                        >                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               <!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 e
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -G-    G-;-    ;-J-    J-}-d    }-/-    /-e7-0    e7--    -b-H	    b-c-P
    c-1-
    1-K׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               H!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 <	         =	  w
         x
                        1         2  l         m                                                M  5            j                                              .   !          O!  '"          p"  H#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q   w  operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       \            B         C                                   N         O                        
         
  Z         [                        #         $  f         g                        /         0  r         s                        ;         <  ~                                 G         H                          "         2  3         3  5         5  P6         Q6  7         7  8         8  :         :  \;         ];  <         <  =         =  %?         &?  h@         i@  A         A  B         B  1D         2D  tE         uE  F         F  G         G  =I         >I  J         J  K         K  M         M  IN         JN  O         O  P         P  R         R  US         VS  T         T  U         U  W         l  n         n  Yo         Zo  p         p  q         q  s         !s  ^t         ct  u         u  v         v  &x         +x  iy         ny  z         z  {         {  2}         7}  u~         z~                         >         C             Ą         Ʉ             J         O                        Չ  Ί                      [  T                                v                        :                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               L!         l!  2         "2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 [
  {                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-;-x    ;-/J-|    /J-}}-\    }}-/-    /--P    -b-    b-c-	    c-=-D
    =-K׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               A!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 K	         L	  
         
                                   I         J                                                           A  &          l  Q            r                                      !          "  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            ?         C                                   K         O                        
         
  W         [                                  $  c         g                        ,         0  o         s                        8         <  {                                 D         H                          "          *  	+          P+  =,          ,  q-          -  .          .  /         2  3         3  
5         5  M6         Q6  7         7  8         8  :         :  Y;         ];  <         <  =         =  "?         &?  e@         i@  A         A  B         B  .D         2D  qE         uE  F         F  G         G  :I         >I  }J         J  K         K  M         M  FN         JN  O         O  P         P  R         R  RS         VS  T         T  U         U  W         l  n         n  Vo         Zo  p         p  q         q  s         !s  at         ct  u         u  v         v  )x         +x  ly         ny  z         z  {         {  5}         7}  x~         z~                         A         C             Ǆ         Ʉ  
           M         O             Ӊ         Չ             Y         [             ߎ           "         v                        =                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               <!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 e
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     - -`      -%ְ-     %ְ-Oܰ-    Oܰ-4-x    4-@-\    @-h-<    h-/-    /-5-    5--    -b- 	    b-c-(
    c-1-
    1-K׳-
                   =         b      J      7   U   7   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               K!         R!  X1                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 64 })  outputElements  5184    input0Elements  31104   input1Elements  24576   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     4                           #         b      J      7   M     operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       @         l                                   W	         ,  f         g                                              M  6            k                                              0            R  *                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     4                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  1728    input0Elements  1728    invocations     1                           3         b      J      7   Q   w  operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       \            A         C                        
           M         O                        
         
  Y         [                        "         $  e         g                        .         0  q         s                        :         <  }                                 F         H                          "         2  3         3  5         5  O6         Q6  7         7  8         8  :         :  [;         ];  <         <  =         =  $?         &?  g@         i@  A         A  B         B  0D         2D  sE         uE  F         F  G         G  <I         >I  J         J  K         K  M         M  HN         JN  O         O  P         P  R         R  TS         VS  T         T  U         U  W         l  n         n  Xo         Zo  p         p  q         q  s         !s  _t         ct  u         u  v         v  'x         +x  jy         ny  z         z  {         {  3}         7}  v~         z~                         ?         C             ń         Ʉ             K         O             щ         Չ             W         [             ݎ            ڏ          v  o                                                      b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     4                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               J!         h!  1         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 [
  {                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     -/-    /-S;-    S;-I-    I-;}-l    ;}--    --P    --    -b-h	    b-o-p
    o-}-
    }-簳-    簳-E׳-L                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ?!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             :         ;  u         v                        &            l                                                          +            R  *          y  Q            x                                             9          :   t!         u!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        %  F&         K&  '         '  (         (  *         *  R+         W+  ,         ,  -         -  /          /  ^0         c0  1          1  2          2  3          ,4  %5          o5  h6          6  7         7  39         89  v:         {:  ;         ;  <         =  ?>         D>  ?         ?  @         @  B         B  KC         PC  D         D  E         E  G         G  WH         \H  I         I  J         J   L         %L  cM         hM  N         N  O         O  ,Q         1Q  oR         tR  S         S  T         T  8V         =V  {W         W  X         X  Z         Z  D[         I[  \         \  ]         ]  _         _  P`         U`  a         a  b         b  d         d  \e         ae  f         f  g         g  %i         *i  hj         mj  k         k  l         l  5n         6n  xo         yo  p         p  q         q  As         Bs  t         t  u         u  
w         w  Mx         Nx  y         y  z         z  |         |  Y}         Z}  ~         ~             "         #  e         f                        .         /  q         r                        :         ;  }         ~                        F         G             ̒         ͒             N         S                                b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               G!         Z!  1         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
                                           b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     6
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -M-    M-A;-x    A;-I-|    I-)}-\    )}--    --h    -b-    b-n-	    n--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               J!         X!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             :         ;  u         v                        &                                              %            F            g  ?            `                                             u                        %          &   `!         a!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ;         <  w         x                        +         ,  g         h             	         	             W         X                        
           F         H                                   6         8  r         t                        &         (  b         d                                     R!         T!  "         "  #         #  %          >%  !&          h&  K'          '  u(          (  )          )  *         :H  uI         vI  J         J  K         K  )M         *M  eN         fN  O         O  P         P  R         R  US         VS  T         T  U         U  	W         
W  EX         FX  Y         Y  Z         Z  [         [  5]         6]  q^         r^  _         _  `         `  %b         &b  ac         bc  d         d  e         e  g         g  Qh         Rh  i         i  j         j  l         l  Am         Bm  |n         ~n  o         o  p         p  0r         2r  ls         ns  t         t  u         u   w         "w  \x         ^x  y         y  z         z  |         |  L}         N}  ~         ~                         <         >  x         z                        ,         .  h         j                                   X         Z             Џ         ҏ             H                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               K!         b!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     1
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     -)-    )-Y;-    Y;-I-    I-A}-l    A}--    --P    --    -b-h	    b-o-p
    o-}-
    }-簳-    簳-E׳-L                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               =!         B!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             :         ;  u         v                        &            k                                                          +            S  +          {  S            {                                             >          ?   y!         z!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        %  G&         K&  '         '  (         (  *         *  S+         W+  ,         ,  -         -  /          /  _0         c0  1         1  2         2  (4         ,4  k5         o5  6         6  7          7  8          89  1:          {:  t;          ;  <          =  =         D>  ?         ?  @         @  	B         B  LC         PC  D         D  E         E  G         G  XH         \H  I         I  J         J  !L         %L  dM         hM  N         N  O         O  -Q         1Q  pR         tR  S         S  T         T  9V         =V  |W         W  X         X  Z         Z  E[         I[  \         \  ]         ]  _         _  Q`         U`  a         a  b         b  d         d  ]e         ae  f         f  g         g  &i         *i  ij         mj  k         k  l         l  4n         6n  wo         yo  p         p  q         q  @s         Bs  t         t  u         u  	w         w  Lx         Nx  y         y  z         z  |         |  X}         Z}  ~         ~             !         #  d         f                        -         /  p         r                        9         ;  |         ~                        E         G             ˒         ͒             O         S                                b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               E!         Z!  1         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
                                           b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     5
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -S-    S-G;-x    G;-I-|    I-/}-\    /}--    --h    -b-    b-n-	    n--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               L!         X!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             :         ;  u         v                        &                                              (             J  "          l  D            f                                                                     /          0   j!         k!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            :         <  v         x                        *         ,  f         h             	         	             V         X                                   G         H                                   7         8  s         t                        '         (  c         d                                     S!         T!  "         "  #         #  %          +  +          :,  -          d-  G.          .  q/          /  0         :H  tI         vI  J         J  K         K  (M         *M  dN         fN  O         O  P         P  R         R  TS         VS  T         T  U         U  W         
W  DX         FX  Y         Y  Z         Z  [         [  4]         6]  p^         r^  _         _  `         `  $b         &b  `c         bc  d         d  e         e  g         g  Ph         Rh  i         i  j         j  l         l  @m         Bm  }n         ~n  o         o  p         p  1r         2r  ms         ns  t         t  u         u  !w         "w  ]x         ^x  y         y  z         z  |         |  M}         N}  ~         ~                        =         >  y         z                        -         .  i         j                                   Y         Z             я         ҏ             I                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               O!         f!  2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     2
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     -)-    )-6-    6-[E-    [E-x-l    x-{9-    {9--x    -b-	    b-i-
    i-w-
    w--|    -ֳ-    ֳ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               =!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         '            N  &          u  M            t                                  -  	          b	  D
          
  y                      A         B  |         }                        6         7  q         r                        "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                       
          T  H                                 D%  &         I  J         J  4L         5L  wM         xM  N         N  O         O  @Q         AQ  R         R  S         S  	U         
U  LV         MV  W         W  X         X  Z         Z  X[         Y[  \         \  ]         ]  !_         "_  d`         e`  a         a  b         b  -d         .d  pe         qe  f         f  g         g  9i         :i  |j         }j  k         k  m         m  An         Fn  o         o  p         p  r         r  Qs         Rs  t         t  u         u  w         w  ]x         ^x  y         y  z         z  &|         '|  i}         j}  ~         ~             2         3  u         v                        >         ?             Ĉ         ň             J         K             Ѝ         э             V         W             ܒ         ݒ              ^         c                        +         ,  j         o                        3         8  v         {                        ?         D             ť         ʥ             K         P             Ѫ         ֪             W         \             ݯ                     %  c         h                        ,         1  o         t                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               <!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
                   
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            0     -,-X     ,--     -;-    ;-C6-x    C6-D-|    D-+x-\    +x-w-    w--h    -b-    b-n-	    n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               B!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         !            B            c  ;            \                                              &	  
          Q
  /         |                        ,         -  g         h           z                        *         +  e         f                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            =         >  {         |                        5         6  s         t             	         	  -         .  k         l                        %         &  c         d                                   [         \                                   S         T                                   K          L   !         !  "         "  $         $  C%          1  1          &2  2          H3   4          j4  B5          5  d6         G  H         H  0J         2J  nK         pK  L         L  M         M  (O         *O  fP         hP  Q         Q  R         R   T         "T  ^U         `U  V         V  W         W  Y         Y  VZ         XZ  [         [  \         \  ^         ^  N_         P_  `         `  a         a  c         
c  Fd         Hd  e         e  f         f   h         h  >i         @i  |j         ~j  k         k  l         n  o         o  *q         ,q  hr         jr  s         s  t         t  "v         $v  `w         bw  x         x  y         y  {         {  X|         Z|  }         }  ~         ~             P         R             ̃         ΃  
           H         J             Ĉ         ƈ             @         B  ~                                 8         :  v         x                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               @!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     1
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     -)-    )-6-    6-aE-    aE-x-l    x-{9-    {9--x    -{b-	    {b-i-
    i-w-
    w--|    -ֳ-    ֳ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ;!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         (             P  (          x  P            x                                  2  	          g	  J
          
                        F         G                                   ;         <  v         w                        '                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                      L   @!          !  ~"          "  #          $  $         D%  &         I  J         J  3L         5L  vM         xM  N         N  O         O  ?Q         AQ  R         R  S         S  U         
U  KV         MV  W         W  X         X  Z         Z  W[         Y[  \         \  ]         ]   _         "_  c`         e`  a         a  b         b  ,d         .d  oe         qe  f         f  g         g  8i         :i  {j         }j  k         k  m         m  Bn         Fn  o         o  p         p  r         r  Ps         Rs  t         t  u         u  w         w  \x         ^x  y         y  z         z  %|         '|  h}         j}  ~         ~             1         3  t         v                        =         ?             È         ň             I         K             ύ         э             U         W             ے         ݒ              _         c                        *         ,  k         o                        4         8  w         {                        @         D             ƥ         ʥ  	           L         P             Ҫ         ֪             X         \             ޯ           !         %  d         h                        -         1  p         t                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               8!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
                   
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            0     -,-X     ,--     -;-    ;-C6-x    C6-D-|    D-+x-\    +x-w-    w--h    -b-    b-n-	    n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               D!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         "            D            f  >            `                                               +	  

          V
  5                                 1         2  l         m                                   /         0  j         k                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            <         >  z         |                        4         6  r         t             	         	  ,         .  j         l                        $         &  b         d                                   Z         \                                   R         T                                   J          L   !         !  "         "  $         $  B%          6  7          7  8          8  9          :  :          6;  <         G  H         H  1J         2J  oK         pK  L         L  M         M  )O         *O  gP         hP  Q         Q  R         R  !T         "T  _U         `U  V         V  W         W  Y         Y  WZ         XZ  [         [  \         \  ^         ^  O_         P_  `         `  a         a  	c         
c  Gd         Hd  e         e  f         f  h         h  ?i         @i  }j         ~j  k         k  l         n  o         o  +q         ,q  ir         jr  s         s  t         t  #v         $v  aw         bw  x         x  y         y  {         {  Y|         Z|  }         }  ~         ~             Q         R             ̓         ΃             I         J             ň         ƈ             A         B                                   9         :  w         x                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               B!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     2
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-M;-x    M;-I-|    I-5}-\    5}-7-    7--h    -b-    b-	n-	    	n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               <!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             :         ;  u         v                        &            l                                                          +            R  *          y  Q            x                                             9          :   t!         u!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            >         @  ~                                  >         @  ~                      	          
  >         @  ~                                  >         @  ~                                  >         @  ~                                  >         @  ~                                  >         @  ~             !         !  "          #  >$         @$  ~%          u&  M'          '  u(          (  )          )  *          +  +         K  L         L  /N         0N  oO         pO  P         P  Q         Q  /S         0S  oT         pT  U         U  V         V  /X         0X  oY         pY  Z         Z  [         [  /]         0]  o^         p^  _         _  `         `  /b         0b  oc         pc  d         d  e         e  /g         0g  oh         ph  i         i  j         j  /l         0l  om         pm  n         n  o         o  /q         0q  or           2         4  r         t                        2         4  r         t                        2         4  r         t                        2         4  r         t                        2         4  r         t                        2         4  r         t                        2         4  r         t                        2                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ?!         R!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     +
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -;-    ;-;-x    ;-}J-|    }J-}-\    }--    --h    -b-    b-kx-	    kx--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               @!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 <	         =	  w
         x
                        1         2  l         m                                                M  5            j                                              .   !          O!  '"          p"  H#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ?         @                                    ?         @                        	          
  ?         @                                    ?         @                                    ?         @                                    ?         @                                    ?         @               !         !  "          #  ?$         @$  %         %  &         &  '          (  ?)         @)  *         *  +         +  ,          -  ?.         @.  /         /  0         0  1          2  ?3         @3  4         4  5         5  6          7  ?8         @8  9         9  :         :  ;          <  ?=         @=  >         >  ?         ?  @          A  ?B         @B  C         C  D         D  E          F  ?G         @G  H         H  I         I  J          K  >L         @L  ~M         M  N         N  O          P  >Q         @Q  ~R         R  S         S  T          U  >V         @V  ~W         W  X         X  Y          Z  >[         @[  ~\         \  ]         ]  ^          _  >`         @`  ~a         a  b         b  c          d  >e         @e  ~f         f  g         g  h          i  >j         @j  ~k         k  l         l  m            x          ň                        K  A            
            ^         `                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               G!         V!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 i
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-S;-x    S;-I-|    I-;}-\    ;}-7-    7--h    -b-    b-	n-	    	n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               <!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             :         ;  u         v                        &            k                                                          +            S  +          {  S            {                                             >          ?   y!         z!  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ?         @                                    ?         @                        	          
  ?         @                                    ?         @                                    ?         @                                    ?         @                                    ?         @               !         !  "          #  ?$         @$  %          =,  -          e-  =.          .  e/          /  0          0  1         K  L         L  .N         0N  nO         pO  P         P  Q         Q  .S         0S  nT         pT  U         U  V         V  .X         0X  nY         pY  Z         Z  [         [  .]         0]  n^         p^  _         _  `         `  .b         0b  nc         pc  d         d  e         e  .g         0g  nh         ph  i         i  j         j  .l         0l  nm         pm  n         n  o         o  .q         0q  nr           3         4  s         t                        3         4  s         t                        3         4  s         t                        3         4  s         t                        3         4  s         t                        3         4  s         t                        3         4  s         t                        3                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ?!         R!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     *
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-;-x    ;-}J-|    }J-}-\    }-]-    ]--h    -b-    b-kx-	    kx--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               >!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 7	         8	  r
         s
                        ,         -  g         h                                               H  1          }  f                                  	            +   !          M!  %"          o"  G#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            >         @  ~                                  >         @  ~                      	          
  >         @  ~                                  >         @  ~                                  >         @  ~                                  >         @  ~                                  >         @  ~             !         !  "          #  >$         @$  ~%         %  &         &  '          (  >)         @)  ~*         *  +         +  ,          -  >.         @.  ~/         /  0         0  1          2  >3         @3  ~4         4  5         5  6          7  >8         @8  ~9         9  :         :  ;          <  >=         @=  ~>         >  ?         ?  @          A  >B         @B  ~C         C  D         D  E          F  >G         @G  ~H         H  I         I  J          K  ?L         @L  M         M  N         N  O          P  ?Q         @Q  R         R  S         S  T          U  ?V         @V  W         W  X         X  Y          Z  ?[         @[  \         \  ]         ]  ^          _  ?`         @`  a         a  b         b  c          d  ?e         @e  f         f  g         g  h          i  ?j         @j  k         k  l         l  m          W  M                      ݒ  ӓ            _         `                          ߘ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               I!         Z!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 i
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-6-x    6-UE-|    UE-x-\    x-$-    $--h    -b-    b-	n-	    	n--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               <!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         '            N  &          u  M            t                                  -  	          b	  D
          
  y                      A         B  |         }                        6         7  q         r                        "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            <         >  z         |                        4         6  r         t             	         	  ,         .  j         l                        $         &  b         d                                   Z         \                                   R         T                                   J          L   !         !  "         "  $         $  B%         H  !J         "J  _K         `K  L         L  M         M  O         O  WP         XP  Q         Q  R         R  T         T  OU         PU  V         V  W         W  	Y         
Y  GZ         HZ  [         [  \         \  ^         ^  ?_         @_  }`         ~`  a         a  b         b  7d         8d  ue         ve  f         f  g         g  /i         0i  mj         nj  k         k  l         l  'n         o  q         q  Er         Fr  s         s  t         t  u          v  =w         >w  {x         |x  y         y  z         z  5|         6|  s}         t}  ~         ~             -         .  k         l                        %         &  c         d             ߉                      [         \             ׎         ؎             S         T             ϓ         Г              b                      ܡ                        0  
                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               >!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     +
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-76-x    76-D-|    D-x-\    x-s-    s--h    -b-    b-ex-	    ex--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               @!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         !            B            c  ;            \                                              &	  
          Q
  /         |                        ,         -  g         h           z                        *         +  e         f                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                        C  9            |                                 p&  '         '  (         (  8*         9*  {+         |+  ,         ,  .         .  D/         E/  0         0  1         1  3         3  P4         Q4  5         5  6         6  8         8  \9         ]9  :         :  ;         ;  %=         &=  h>         i>  ?         ?  @         @  1B         2B  tC         uC  D         D  E         E  =G         >G  H         H  I         I  K         K  IL         JL  M         M  N         N  P         P  QQ         VQ  R         R  S         S  U         U  ]V         bV  W         W  X         X  &Z         +Z  i[         n[  \         \  ]         ]  2_         7_  u`         z`  a         a  b          c  >d         Cd  e         e  f         f  h         h  Ji         Oi  j         j  k         k  m         m  Vn         [n  o         o  p         p  r         $r  bs         gs  t         t  u         u  +w         0w  nx         sx  y         y  z         z  7|         <|  z}         }  ~         ~              C         H             Ƀ         ΃             O         T             Ո         ڈ             [         `                        $         )  g         l                        0         5  s         x                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               >!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 i
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-6-x    6-[E-|    [E-x-\    x-W*-    W*--h    -{b-    {b-n-	    n--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               <!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         (             P  (          x  P            x                                  2  	          g	  J
          
                        F         G                                   ;         <  v         w                        '                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            =         >  {         |                        5         6  s         t             	         	  -         .  k         l                        %         &  c         d                                   [         \                                   S         T                                   K          L   !         !  "         "  $         $  C%         H   J         "J  ^K         `K  L         L  M         M  O         O  VP         XP  Q         Q  R         R  T         T  NU         PU  V         V  W         W  Y         
Y  FZ         HZ  [         [  \         \   ^         ^  >_         @_  |`         ~`  a         a  b         b  6d         8d  te         ve  f         f  g         g  .i         0i  lj         nj  k         k  l         l  &n         o  q         q  Dr         Fr  s         s  t         t  u          v  <w         >w  zx         |x  y         y  z         z  4|         6|  r}         t}  ~         ~             ,         .  j         l                        $         &  b         d             މ                      Z         \             ֎         ؎             R         T             Γ         Г            Z  4            ^                      ب              ܪ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               <!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     *
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-=6-x    =6-D-|    D-%x-\    %x-y-    y--h    -b-    b-ex-	    ex--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               >!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         "            D            f  >            `                                               +	  

          V
  5                                 1         2  l         m                                   /         0  j         k                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         O  E                        	          
            [  Q         p&  '         '  (         (  7*         9*  z+         |+  ,         ,   .         .  C/         E/  0         0  1         1  3         3  O4         Q4  5         5  6         6  8         8  [9         ]9  :         :  ;         ;  $=         &=  g>         i>  ?         ?  @         @  0B         2B  sC         uC  D         D  E         E  <G         >G  H         H  I         I  K         K  HL         JL  M         M  N         N  P         P  RQ         VQ  R         R  S         S  U         U  ^V         bV  W         W  X         X  'Z         +Z  j[         n[  \         \  ]         ]  3_         7_  v`         z`  a         a  b          c  ?d         Cd  e         e  f         f  h         h  Ki         Oi  j         j  k         k  m         m  Wn         [n  o         o  p         p   r         $r  cs         gs  t         t  u         u  ,w         0w  ox         sx  y         y  z         z  8|         <|  {}         }  ~         ~             D         H             ʃ         ΃             P         T             ֈ         ڈ             \         `                        %         )  h         l                        1         5  t         x                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               >!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 i
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-;-x    ;-J-|    J-k}-\    k}--    --h    -cb-    cb-Yx-	    Yx--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               @!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 K	         L	  
         
                                   I         J                                                           A  %          l  P            s                                      !          "  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            :         <  v         x                        *         ,  f         h             	         	             V         X                        
           F         H                                   6         8  r         t                        &         (  b         d                                     R!         T!  "         "  #         #  %         %  B&         D&  ~'         '  (         (  )         )  2+         4+  n,         p,  -         -  .         .  "0         $0  ^1         `1  2         2  3         3  5         5  N6         P6  7         7  8         8  :         :  >;         @;  z<         |<  =         =  >         >  .@         0@  jA         lA  B         B  C         C  E          E  ZF         \F  G         G  H         H  J         J  WK         XK  L         L  M         M  O         O  GP         HP  Q         Q  R         R  S         S  7U         8U  sV         tV  W         W  X         X  'Z         (Z  c[         d[  \         \  ]         ]  _         _  S`         T`  a         a  b         b  d         d  Ce         De  f         f  g         g  h         h  3j         4j  ok         pk  l         l  m         m  #o          `  \                                  )  %          l  h                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               9!         :!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 _
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-y-     y--    -;-x    ;-wJ-|    wJ-}-\    }-7-    7--h    -b-    b--	    --h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               8!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 <	         =	  w
         x
                        1         2  l         m                                                M  5            j                                              .   !          O!  '"          p"  H#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            B         C                                   N         O                        
         
  Z         [                        #         $  f         g                        /         0  r         s                        ;         <  ~                                 G         H                          "         "  S#         T#  $         $  %         %  '         '  _(         `(  )         )  *         *  (,         ),  k-         l-  .         .  /         /  41         51  w2         x2  3         3  4         4  @6         A6  7         7  8         8  	:         
:  L;         M;  <         <  =         =  ?         ?  X@         Y@  A         A  B         B  !D         "D  dE         eE  F         F  G         G  -I         .I  pJ         qJ  K          X  X          (Y   Z          JZ  "[          l[  D\          \  f]         n  o         o  !q         &q  dr         	  G         L             ͙         ҙ             S         X             ٞ         ޞ           !  _         d                        (         -  k         p                        4         9  w         |                        @         E             Ʊ         ˱  	           L         Q             Ҷ         ׶                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               O!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-;-x    ;-J-|    J-k}-\    k}-1-    1--h    -ob-    ob-Yx-	    Yx--h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               >!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 K	         L	  
         
                                   I         J                                                           A  &          l  Q            r                                      !          "  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ;         <  w         x                        +         ,  g         h             	         	             W         X                                   G         H                                   7         8  s         t                        '         (  c         d                                     S!         T!  "         "  #         #  %         %  C&         D&  '         '  (         (  )         )  3+         4+  o,         p,  -         -  .         .  #0         $0  _1         `1  2         2  3         3  5         5  O6         P6  7         7  8         8  :         :  ?;         @;  {<         |<  =         =  >         >  /@         0@  kA         lA  B         B  C         C  E          E  [F         \F  G         G  H         H  J         J  VK         XK  L         L  M         M  
O         O  FP         HP  Q         Q  R         R  S         S  6U         8U  rV         tV  W         W  X         X  &Z         (Z  b[         d[  \         \  ]         ]  _         _  R`         T`  a         a  b         b  d         d  Be         De  ~f         f  g         g  h         h  2j         4j  nk         pk  l         l  m         m  "o                                  5  1          x  t                               >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               =!         >!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 _
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     --    -;-x    ;-wJ-|    wJ-}-\    }-7-    7--h    -b-    b--	    --h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               :!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 7	         8	  r
         s
                        ,         -  g         h                                               H  1          }  f                                  	            +   !          M!  %"          o"  G#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            A         C                        
           M         O                        
         
  Y         [                        "         $  e         g                        .         0  q         s                        :         <  }                                 F         H                          "         "  R#         T#  $         $  %         %  '         '  ^(         `(  )         )  *         *  ',         ),  j-         l-  .         .  /         /  31         51  v2         x2  3         3  4         4  ?6         A6  7         7  8         8  :         
:  K;         M;  <         <  =         =  ?         ?  W@         Y@  A         A  B         B   D         "D  cE         eE  F         F  G         G  ,I         .I  oJ         qJ  K          ]  ^          ^  _          _  `          a  a          8b  c         n  o         o  "q         &q  er         	  H         L             Ι         ҙ             T         X             ڞ         ޞ           !  `         d                        )         -  l         p                        5         9  x         |                        A         E             Ǳ         ˱  
           M         Q             Ӷ         ׶                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               K!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-C-     C--    -6-x    6-+E-|    +E-yx-\    yx--    --h    -b-    b-Sx-	    Sx--h
    -%-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     L                       T!  1          1  z=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         !            B            c  ;            \                                              >	  $
          q
  W                                 Z         [                                   M         N                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                        C  9            |                                 &  '         '  )         ")  `*         e*  +         +  ,         ,  ).         ..  l/         q/  0         0  1         1  53         :3  x4         }4  5         5  6         7  A8         F8  9         9  :         :  
<         <  M=         R=  >         >  ?         ?  A         A  YB         ^B  C         C  D         D  "F         'F  eG         jG  H         H  I         I  .K         3K  qL         vL  M         M  N         N  >P         ?P  Q         Q  R         R  T         T  JU         KU  V         V  W         W  Y         Y  VZ         WZ  [         [  \         \  ^          ^  b_         c_  `         `  a         a  +c         ,c  nd         od  e         e  f         f  7h         8h  zi         {i  j         j   l         l  Cm         Dm  n         n  o         o  q         q  Or         Pr  s         s  t         t  v         v  [w         \w  x         x  y         y  ${         %{  g|         h|  }         }  ~         ~  0         1  s         t                        <         =                      Ç             H         I             Ό         ό             T         U             ڑ         ۑ             `         a                        )                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               K!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 a
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -{,-X     {,-1-     1--    -C6-x    C6-D-|    D-+x-\    +x--    --h    -Wb-    Wb-Yx-	    Yx--h
    -%-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     L                       H!  1          1  h=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     :            W  1          t  N            w                                  	  	          E
  +         r                        (         )  e         f                                   :         ;  w         x                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                        :  -          t  g                                 &  '         '  !)         ")  d*         e*  +         +  ,         ,  -.         ..  p/         q/  0         0  1         1  93         :3  |4         }4  5         5  7         7  E8         F8  9         9  :         :  <         <  Q=         R=  >         >  ?         ?  A         A  ]B         ^B  C         C  D         D  &F         'F  iG         jG  H         H  I         I  2K         3K  uL         vL  M         M  N         N  :P         ?P  }Q         Q  R         R  T         T  FU         KU  V         V  W         W  Y         Y  RZ         WZ  [         [  \         \  ^          ^  ^_         c_  `         `  a         a  'c         ,c  jd         od  e         e  f         f  3h         8h  vi         {i  j         j  k         l  ?m         Dm  n         n  o         o  q         q  Kr         Pr  s         s  t         t  v         v  Ww         \w  x         x  y         y   {         %{  c|         h|  }         }  ~         ~  ,         1  o         t                        8         =  {                    Ç             D         I             ʌ         ό             P         U             ֑         ۑ             \         a                        %                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               7!         8!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 g
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,-=-     =--    -6-x    6-1E-|    1E-x-\    x--    --h    -b-    b-Sx-	    Sx--h
    -%-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     K                       P!  1          1  s=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         "            D            f  >            `                                              C	  *
          v
  ]                      "         #  _         `                                   R         S                        	                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         O  E                        	          
            [  Q         &  '         '  )         ")  a*         e*  +         +  ,         ,  *.         ..  m/         q/  0         0  1         1  63         :3  y4         }4  5         5  6         7  B8         F8  9         9  :         :  <         <  N=         R=  >         >  ?         ?  A         A  ZB         ^B  C         C  D         D  #F         'F  fG         jG  H         H  I         I  /K         3K  rL         vL  M         M  N         N  =P         ?P  Q         Q  R         R  T         T  IU         KU  V         V  W         W  Y         Y  UZ         WZ  [         [  \         \  ^          ^  a_         c_  `         `  a         a  *c         ,c  md         od  e         e  f         f  6h         8h  yi         {i  j         j  k         l  Bm         Dm  n         n  o         o  q         q  Nr         Pr  s         s  t         t  v         v  Zw         \w  x         x  y         y  #{         %{  f|         h|  }         }  ~         ~  /         1  r         t                        ;         =  ~                    Ç             G         I             ͌         ό             S         U             ّ         ۑ             _         a                        (                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               K!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 a
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -u,-X     u,-+-     +--    -C6-x    C6-D-|    D-+x-\    +x--    --h    -cb-    cb-Yx-	    Yx--h
    -%-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     K                       D!  1          1  a=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     :            W  0          t  M            x                                  	  	          E
  ,         r                        (         )  e         f                                   :         ;  w         x                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         "            \  O            	          	  
          
           &  '         '   )         ")  c*         e*  +         +  ,         ,  ,.         ..  o/         q/  0         0  1         1  83         :3  {4         }4  5         5  7         7  D8         F8  9         9  :         :  <         <  P=         R=  >         >  ?         ?  A         A  \B         ^B  C         C  D         D  %F         'F  hG         jG  H         H  I         I  1K         3K  tL         vL  M         M  N         N  ;P         ?P  ~Q         Q  R         R  T         T  GU         KU  V         V  W         W  Y         Y  SZ         WZ  [         [  \         \  ^          ^  __         c_  `         `  a         a  (c         ,c  kd         od  e         e  f         f  4h         8h  wi         {i  j         j  k         l  @m         Dm  n         n  o         o  	q         q  Lr         Pr  s         s  t         t  v         v  Xw         \w  x         x  y         y  !{         %{  d|         h|  }         }  ~         ~  -         1  p         t                        9         =  |                    Ç             E         I             ˌ         ό             Q         U             ב         ۑ             ]         a                        &                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ;!         <!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 g
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -u,-X     u,-+-     +--    -;-x    ;-MJ-|    MJ-}-\    }-9-    9--h    -b-    b--	    --h
    -+-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     H                       D!  1          1  ^=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      
           G	         H	  
         
                                   Y         Z                                               .  
          [  7            d                                               $!  !          I"  !#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            >         C                                   J         O                        
         
  V         [                                 $  b         g                        +         0  n         s                        7         <  z                                  C         H                          "         "  O#         T#  $         $  %         %  '         '  [(         `(  )         )  *         *  $,         ),  g-         l-  .         .  /         /  01         51  s2         x2  3         3  4         4  <6         A6  7         7  8         8  :         
:  H;         M;  <         <  =         =  ?         ?  T@         Y@  A         A  B         B  D         "D  `E         eE  F         F  G         G  )I         .I  lJ         qJ  K          >X  Y          dY  <Z          Z  b[          [  \          \  ]         Z             ߖ           "         #  e         f                        .         /  q         r                        :         ;  }         ~                        F         G             ̩         ͩ             R         S             خ         ٮ             ^         _                        '         (  j         k                        3                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               J!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -W,-X     W,--     --    -;-x    ;-J-|    J-}-\    }-9-    9--h    -ob-    ob--	    --h
    -+-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     E                       4!  y1          1  C=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 8	         9	  u
         v
                        +         ,  h         i                                               I  %          |  X                                              8   !          c!  ;"          "  f#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            B         C                                   N         O                        
         
  Z         [                        #         $  f         g                        /         0  r         s                        ;         <  ~                                 G         H                          "         "  S#         T#  $         $  %         %  '         '  _(         `(  )         )  *         *  (,         ),  k-         l-  .         .  /         /  41         51  w2         x2  3         3  4         4  @6         A6  7         7  8         8  	:         
:  L;         M;  <         <  =         =  ?         ?  X@         Y@  A         A  B         B  !D         "D  dE         eE  F         F  G         G  -I         .I  pJ         qJ  K          W  XX          X  Y          Y  Z          [  [          0\  ]         Z             ۖ                    #  a         f                        *         /  m         r                        6         ;  y         ~                        B         G             ȩ         ͩ             N         S             Ԯ         ٮ             Z         _                        #         (  f         k                        /                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               :!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -{,-X     {,-1-     1--    -;-x    ;-SJ-|    SJ-}-\    }-9-    9--h    -b-    b--	    --h
    -+-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     I                       H!  1          1  e=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      
           G	         H	  
         
                                   Y         Z                                               .            [  8            e                                                '!  !          M"  %#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ?         C                                   K         O                        
         
  W         [                                  $  c         g                        ,         0  o         s                        8         <  {                                 D         H                          "         "  P#         T#  $         $  %         %  '         '  \(         `(  )         )  *         *  %,         ),  h-         l-  .         .  /         /  11         51  t2         x2  3         3  4         4  =6         A6  7         7  8         8  :         
:  I;         M;  <         <  =         =  ?         ?  U@         Y@  A         A  B         B  D         "D  aE         eE  F         F  G         G  *I         .I  mJ         qJ  K          ]  ^          "_  _          H`   a          na  Fb          b  lc         Z             ޖ           !         #  d         f                        -         /  p         r                        9         ;  |         ~                        E         G             ˩         ͩ             Q         S             ׮         ٮ             ]         _                        &         (  i         k                        2                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               F!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -W,-X     W,--     --    -;-x    ;-J-|    J-}-\    }-9-    9--h    -ib-    ib--	    --h
    -+-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     D                       2!  v1          1  ?=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         }                        3	         4	  p
         q
                        &         '  c         d                                               D  !          w  T                                  	            5   !          a!  9"          "  e#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            A         C                        
           M         O                        
         
  Y         [                        "         $  e         g                        .         0  q         s                        :         <  }                                 F         H                          "         "  R#         T#  $         $  %         %  '         '  ^(         `(  )         )  *         *  ',         ),  j-         l-  .         .  /         /  31         51  v2         x2  3         3  4         4  ?6         A6  7         7  8         8  :         
:  K;         M;  <         <  =         =  ?         ?  W@         Y@  A         A  B         B   D         "D  cE         eE  F         F  G         G  ,I         .I  oJ         qJ  K          \]  4^          ^  `_          _  `          `  a          b  b         Z             ܖ                    #  b         f                        +         /  n         r                        7         ;  z         ~                         C         G             ɩ         ͩ             O         S             ծ         ٮ             [         _                        $         (  g         k                        0                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               8!         @!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,--     -/-    /-Y;-x    Y;-I-|    I-A}-\    A}--    -w-h    w-b-    b--	    --h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               A!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             <         =  y         z                        0            |                                              8            ]  5          ~  V            w                                                         I          J   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            B         C                                   N         O                        
         
  Z         [                        #         $  f         g                        /         0  r         s                        ;         <  ~                                 G         H                          "         "  S#         T#  $         $  %         %  '         '  [(         `(  )         )  *         *  $,         ),  g-         l-  .         .  /         /  01         51  s2         x2  3         3  4         4  <6         A6  7         7  8         8  :         
:  H;         M;  <         <  =         =  ?         ?  T@         Y@  A         A  B         B  D         "D  `E         eE  F         F  G         G  )I          x`  Pa          a  vb          b  c          c  d          e  e         l  m         m  o         o  Rp         Sp  q         q  r         r  t         t  ^u         _u  v         v  w         w  'y         (y  jz         kz  {         {  |         |  3~         4~  v         w                        ?         @             Ņ         ƅ           	  K         L             ъ         Ҋ             W         X             ݏ         ޏ            !  c           ؔ         ٔ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     G                       >!  1          1  T=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -#-    #-e;-x    e;-I-|    I-M}-\    M}--    --h    -'b-    'b--	    --h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               =!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             <         =  y         z                        0                                              C            n  F            s                                                                                 S          T   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            >         C                                   J         O                        
         
  V         [                                 $  b         g                        +         0  n         s                        7         <  z                                  C         H                          "         "  O#         T#  $         $  %         %  '         '  _(         `(  )         )  *         *  (,         ),  k-         l-  .         .  /         /  41         51  w2         x2  3         3  4         4  @6         A6  7         7  8         8  	:         
:  L;         M;  <         <  =         =  ?         ?  X@         Y@  A         A  B         B  !D         "D  dE         eE  F         F  G         G  -I          a  b          b  c          ,d  e          ^e  Df          f  vg         l  m         m  o         o  Np         Sp  q         q  r         r  t         t  Zu         _u  v         v  w         w  #y         (y  fz         kz  {         {  |         |  /~         4~  r         w                        ;         @  ~                    ƅ           	  G         L             ͊         Ҋ             S         X             ُ         ޏ           !  _           Ԕ         ٔ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     G                       J!  1          1  f=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-e;-x    e;-I-|    I-M}-\    M}--    -w-h    w-b-    b--	    --h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               A!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             <         =  y         z                        0            |                                              <            b  :            \            ~                                                         S          T   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            A         C                        
           M         O                        
         
  Y         [                        "         $  e         g                        .         0  q         s                        :         <  }                                 F         H                          "         "  R#         T#  $         $  %         %  '         '  \(         `(  )         )  *         *  %,         ),  h-         l-  .         .  /         /  11         51  t2         x2  3         3  4         4  =6         A6  7         7  8         8  :         
:  I;         M;  <         <  =         =  ?         ?  U@         Y@  A         A  B         B  D         "D  aE         eE  F         F  G         G  *I          6f  g          \g  4h          h  Zi          i  j          j  k         l  m         m  o         o  Qp         Sp  q         q  r         r  t         t  ]u         _u  v         v  w         w  &y         (y  iz         kz  {         {  |         |  2~         4~  u         w                        >         @             ą         ƅ           	  J         L             Њ         Ҋ             V         X             ܏         ޏ           !  b           ה         ٔ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     F                       >!  1          1  S=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -)-    )-e;-x    e;-I-|    I-M}-\    M}--    --h    -'b-    'b--	    --h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ?!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             <         =  y         z                        0                                              F            r  J            w                                                                                 X          Y   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ?         C                                   K         O                        
         
  W         [                                  $  c         g                        ,         0  o         s                        8         <  {                                 D         H                          "         "  P#         T#  $         $  %         %  '         '  ^(         `(  )         )  *         *  ',         ),  j-         l-  .         .  /         /  31         51  v2         x2  3         3  4         4  ?6         A6  7         7  8         8  :         
:  K;         M;  <         <  =         =  ?         ?  W@         Y@  A         A  B         B   D         "D  cE         eE  F         F  G         G  ,I          g  h          h  i          &j  k          Xk  >l         l  m         m  o         o  Op         Sp  q         q  r         r  t         t  [u         _u  v         v  w         w  $y         (y  gz         kz  {         {  |         |  0~         4~  s         w                        <         @                      ƅ           	  H         L             Ί         Ҋ             T         X             ڏ         ޏ           !  `          d  J           Ք         ٔ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     F                       J!  1          1  e=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -5-    5-;-    ;-SJ-    SJ-}-d    }-1-    1--0    --    -Kb-H	    Kb-c-P
    c-+-
    +-9׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               B!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      
           G	         H	  
         
                                   Y         Z                                               .  
          [  7            d                                               $!  !          I"  !#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q   w  operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       \            >         C                                   J         O                        
         
  V         [                                 $  b         g                        +         0  n         s                        7         <  z                                  C         H                          "         "  O#         T#  $         $  %         %  '         '  [(         `(  )         )  *         *  $,         ),  g-         l-  .         .  /         /  01         51  s2         x2  3         3  4         4  <6         A6  7         7  8         8  :         
:  H;         M;  <         <  =         =  ?         ?  T@         Y@  A         A  B         B  D          ng  eh          h  i          i  j          7k  .l          zl  qm         m  n          o  >p         Cp  q         q  r         r  t         t  Ju         Ou  v         v  w         w  y         y  Sz         Xz  {         {  |         |  ~         ~  ]         b                        &         +  l         m                        2         3  t         u                        =         >           4  r         w                                b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     M                       \!  1          
2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 e
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    <     -,-p     ,--     -/-    /-O6-    O6-D-    D-7x-t    7x-G4-     G4-;-X    ;--    -b-p	    b-i-x
    i-w-
    w--    -ֳ-\    ֳ--<                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ?!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     :            W  1          t  N            w                                  	  	          E
  +         r                        (         )  e         f                                   :         ;  w         x                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h         (#  $          CF  7G         G  H         H  I         I  8K         :K  vL         xL  M         M  N         N  0P         2P  nQ         pQ  R         R  S         S  (U         *U  fV         hV  W         W  X         X   Z         "Z  ^[         `[  \         \  ]         ]  _         _  V`         X`  a         a  b         b  d         d  Ne         Pe  f         f  g         g  i         
i  Fj         Hj  k         k  l         l   n         n  >o         @o  |p         ~p  q         q  r         r  6t         8t  tu         vu  v         v  w         w  .y         0y  lz         nz  {         {  |         |  &~         (~  d         f                                    \         ^             ؆         چ             T         V             Ћ         ҋ             L         N             Ȑ          ʐ             C          E  9            v                      9         ;  w         y                        1         3  o         q                        )         +  g         i                        !         #  _         a             ۪         ݪ             W         Y             ӯ         կ             O         Q             ˴                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               @!         P!  1         1  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
  	         
                                b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
  	                              b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            4     -,-`     ,--     -5-    5-;-    ;-YJ-    YJ-}-d    }-1-    1--0    --    -Eb-H	    Eb-c-P
    c-+-
    +-9׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               A!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      
           G	         H	  
         
                                   Y         Z                                               .            [  8            e                                                '!  !          M"  %#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q   w  operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       \            ?         C                                   K         O                        
         
  W         [                                  $  c         g                        ,         0  o         s                        8         <  {                                 D         H                          "         "  P#         T#  $         $  %         %  '         '  \(         `(  )         )  *         *  %,         ),  h-         l-  .         .  /         /  11         51  t2         x2  3         3  4         4  =6         A6  7         7  8         8  :         
:  I;         M;  <         <  =         =  ?         ?  U@         Y@  A         A  B         B  D         ng  h         h  i         i  3k         7k  vl         zl  m         m  n          o  ?p          Cp  :q          q  }r          r  s          t  u          Ou  Fv         v  w         w  y         y  Tz         Xz  {         {  |         |  ~         ~  ^         b                        '         +  k         m                        1         3  s         u                        <         >           4  s         w                                b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     L                       X!  1          2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 f
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    <     -,-p     ,--     -5-    5-O6-    O6-D-    D-7x-t    7x-G4-     G4-;-X    ;--    -b-p	    b-i-x
    i-w-
    w--    -ֳ-\    ֳ--<                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               A!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     :            W  0          t  M            x                                  	  	          E
  ,         r                        (         )  e         f                                   :         ;  w         x                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        (#  d$         CF  G         G  H         H  I         I  9K         :K  wL         xL  M         M  N         N  1P         2P  oQ         pQ  R         R  S         S  )U         *U  gV         hV  W         W  X         X  !Z         "Z  _[         `[  \         \  ]         ]  _         _  W`         X`  a         a  b         b  d         d  Oe         Pe  f         f  g         g  	i         
i  Gj         Hj  k         k  l         l  n         n  ?o         @o  }p         ~p  q         q  r         r  7t         8t  uu         vu  v         v  w         w  /y         0y  mz         nz  {         {  |         |  '~         (~  e         f                                    ]         ^             ن         چ             U         V             ы         ҋ             M         N             ɐ         ʐ             D         E                                               ;  0          y  n                       2         3  p         q                        *         +  h         i                        "         #  `         a             ܪ         ݪ             X         Y             ԯ         կ             P         Q             ̴                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               @!         P!  1         1  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
  	         
                                b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
  	                              b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            4     -,-`     ,--     -5-    5-_;-    _;-I-    I-G}-d    G}-%-    %--p    -b-    b-o-
    o-}-
    }-հ-t    հ-?׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ?!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             <         =  y         z                        0            |                                              8            ]  5          ~  V            w                                                         I          J   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |           t                                   @  4            t         d#  $         $  %         %  "'         $'  b(         d(  )         )  *         *  ",         $,  b-         d-  .         .  /         /  "1         $1  b2         d2  3         3  4         4  "6         $6  b7         d7  8         8  9         9  ";         $;  b<         d<  =         =  >         >  "@         $@  bA         dA  B         B  C         C  "E         $E  bF         dF  G         G  H         H  "J         $J  cK         dK  L         L  M         M  #O         $O  cP         dP  Q         Q  R         R  #T         $T  cU         dU  V         V  W         W  #Y         $Y  cZ         dZ  [         [  \         \  #^         $^  c_         d_  `         `  a         a  #c         $c  cd         dd  e         e  f         f  #h         $h  ci         di  j         j  k         k  #m         $m  cn         dn  o         ds  t         t  u         u  "w         $w  bx         dx  y         y  z         z  "|         $|  b}         d}  ~         ~             "         $  b         d                        "         $  b         d                        "         $  b         d                        "         $  b         d                        "         $  b         d                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               K!         f!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
  	         
        
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     *
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-k;-x    k;-J-|    J-S}-\    S}-2-    2--h    -b-    b-n-	    n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               @!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             <         =  y         z                        0                                              C            n  F            s                                                                                 S          T   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ?         @                                    ?         @                        	          
  ?         @                                    ?         @                                    ?         @                                    ?         @                                    ?         @               !          0/  0          ^0  C1          1  q2          2  3          3  4         F  H         H  GI         I  J         J  L         L  QM         SM  N         N  O         O  Q         Q  QR         SR  S         S  T         T  V         V  QW         SW  X         X  Y         Y  [         [  Q\         S\  ]         ]  ^         ^  `         `  Qa         Sa  b         b  c         c  e         e  Qf         Sf  g         g  h         h  j         j  Qk         Sk  l         l  m         m  o         o  p         p  #r         $r  cs           א         ؐ             W         X             ו         ؕ             W         X             ך         ؚ             W         X             ן         ؟             W         X             פ         ؤ             W         X             ש         ة             W         X             ׮         خ             W         X                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               C!         V!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     -
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -5-    5-k;-    k;-J-    J-S}-d    S}-%-    %--p    -b-    b-o-
    o-}-
    }-հ-t    հ-?׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ?!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             <         =  y         z                        0            |                                              <            b  :            \            ~                                                         S          T   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                  @  4            t                     d#  $         $  %         %  #'         $'  c(         d(  )         )  *         *  #,         $,  c-         d-  .         .  /         /  #1         $1  c2         d2  3         3  4         4  #6         $6  c7         d7  8         8  9         9  #;         $;  c<         d<  =         =  >         >  #@         $@  cA         dA  B         B  C         C  #E         $E  cF         dF  G         G  H         H  #J         $J  bK         dK  L         L  M         M  "O         $O  bP         dP  Q         Q  R         R  "T         $T  bU         dU  V         V  W         W  "Y         $Y  bZ         dZ  [         [  \         \  "^         $^  b_         d_  `         `  a         a  "c         $c  bd         dd  e         e  f         f  "h         $h  bi         di  j         j  k         k  "m         $m  bn         dn  o         ds  t         t  u         u  #w         $w  cx         dx  y         y  z         z  #|         $|  c}         d}  ~         ~             #         $  c         d                        #         $  c         d                        #         $  c         d                        #         $  c         d                        #         $  c         d                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               K!         f!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
  	         
        
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     )
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-q;-x    q;-J-|    J-Y}-\    Y}-2-    2--h    -b-    b-n-	    n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               @!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             <         =  y         z                        0                                              F            r  J            w                                                                                 X          Y   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            >         @  ~                                  >         @  ~                      	          
  >         @  ~                                  >         @  ~                                  >         @  ~                                  >         @  ~                                  >         @  ~             !          5  5          D6  )7          r7  W8          8  9          9  :         F  H         H  FI         I  J         J  L         L  RM         SM  N         N  O         O  Q         Q  RR         SR  S         S  T         T  V         V  RW         SW  X         X  Y         Y  [         [  R\         S\  ]         ]  ^         ^  `         `  Ra         Sa  b         b  c         c  e         e  Rf         Sf  g         g  h         h  j         j  Rk         Sk  l         l  m         m  o         o  p         p  "r         $r  bs           ֐         ؐ             V         X             ֕         ؕ             V         X             ֚         ؚ             V         X             ֟         ؟             V         X             ֤         ؤ             V         X             ֩         ة             V         X             ֮         خ             V         X                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               E!         Z!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     .
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     -;-    ;-6-    6-=E-    =E-x-l    x-8-    8--x    -b-	    b-u-
    u--
    --|    -%ֳ-    %ֳ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               A!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         !            B            c  ;            \                                              >	  $
          q
  W                                 Z         [                                   M         N                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         =  ~>          >  ?          ?  @          *A  B          ZB  >C         zI  J         J  K         K  2M         4M  pN         rN  O         O  P         P  *R         ,R  hS         jS  T         T  U         U  "W         $W  `X         bX  Y         Y  Z         Z  \         \  X]         Z]  ^         ^  _         _  a         a  Pb         Rb  c         c  d         d  
f         f  Hg         Jg  h         h  i         i  k         k  Al         Bl  m         m  n         n  o         o  9q         :q  wr         xr  s         s  t         t  1v         2v  ow         pw  x         x  y         y  ){         *{  g|         h|  }         }  ~         ~  !         "  _         `             ۃ         ܃             W         X             ӈ         Ԉ             O         P             ˍ         ̍  	         
  G         H             Ò         Ē             ?         @  }         ~                        7         8  u         v                        /         0  m         n                        '         (  e         f                                    ]         ^             ٬         ڬ             U         V             ѱ         ұ             L         N             ȶ         ޶             Y                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               B!         T!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
  	         
        
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
  	                              b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     f
            0     -,-X     ,--     -5-    5-U6-x    U6-D-|    D-=x-\    =x--    --h    -b-    b-n-	    n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               >!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     :            W  1          t  N            w                                  	  	          E
  +         r                        (         )  e         f                                   :         ;  w         x                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            B         C                                   N         O                        
         
  Z         [                        #         $  f         g                        /         0  r         s                        ;         <  ~                                 G         H                          "         "  S#         T#  $         $  %         %  '         '  [(         `(  )         )  *         *  $,         ),  g-         l-  .         .  /         /  01         51  s2         x2  3         3  4         4  <6         A6  7         7  8         8  :         
:  H;         M;  <         <  =         =  ?         ?  T@         Y@  A         A  B         B  D         "D  `E         eE  F          ]  Z^          ^  {_          _  `          `  a          b  b         i  j         j  7l         <l  zm         m  n         n   p         p  Cq         Hq  r         r  s         s  u         u  Ov         Tv  w         w  x         x  z         z  [{         `{  |         |  }         }  $         )  g         l                        0         5  s         x                        <         A                      ǋ           
  H         M             ΐ         Ӑ             T         Y                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               >!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     -
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     -;-    ;-6-    6-CE-    CE-x-l    x-8-    8--x    -b-	    b-u-
    u--
    --|    -%ֳ-    %ֳ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               A!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         "            D            f  >            `                                              C	  *
          v
  ]                      "         #  _         `                                   R         S                        	                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         C  nD          D  E          E  F          G  G          JH  .I         zI  J         J  K         K  3M         4M  qN         rN  O         O  P         P  +R         ,R  iS         jS  T         T  U         U  #W         $W  aX         bX  Y         Y  Z         Z  \         \  Y]         Z]  ^         ^  _         _  a         a  Qb         Rb  c         c  d         d  f         f  Ig         Jg  h         h  i         i  k         k  @l         Bl  ~m         m  n         n  o         o  8q         :q  vr         xr  s         s  t         t  0v         2v  nw         pw  x         x  y         y  ({         *{  f|         h|  }         }  ~         ~            "  ^         `             ڃ         ܃             V         X             ҈         Ԉ             N         P             ʍ         ̍           
  F         H                      Ē              >         @  |         ~                        6         8  t         v                        .         0  l         n                        &         (  d         f                                    \         ^             ج         ڬ             T         V             б         ұ             M         N             ɶ         ޶             X                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               @!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
  	         
        
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
  	                              b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     e
            0     -,-X     ,--     -5-    5-U6-x    U6-D-|    D-=x-\    =x--    --h    -b-    b-n-	    n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               >!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     :            W  0          t  M            x                                  	  	          E
  ,         r                        (         )  e         f                                   :         ;  w         x                                b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            A         C                        
           M         O                        
         
  Y         [                        "         $  e         g                        .         0  q         s                        :         <  }                                 F         H                          "         "  R#         T#  $         $  %         %  '         '  \(         `(  )         )  *         *  %,         ),  h-         l-  .         .  /         /  11         51  t2         x2  3         3  4         4  =6         A6  7         7  8         8  :         
:  I;         M;  <         <  =         =  ?         ?  U@         Y@  A         A  B         B  D         "D  aE         eE  F          'c  c          Hd   e          ie  Af          f  bg          g  h         i  j         j  8l         <l  {m         m  n         n  p         p  Dq         Hq  r         r  s         s  u         u  Pv         Tv  w         w  x         x  z         z  \{         `{  |         |  }         }  %         )  h         l                        1         5  t         x                        =         A             Ë         ǋ           
  I         M             ϐ         Ӑ             U         Y                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               >!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     .
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -S-    S-;-x    ;-_J-|    _J-}-\    }-i6-    i6--h    -b-    b-kx-	    kx--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               J!         \!  1         
2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      
           G	         H	  
         
                                   Y         Z                                               .  
          [  7            d                                               $!  !          I"  !#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            >         C                                   J         O                        
         
  V         [                                 $  b         g                        +         0  n         s                        7         <  z                                  C         H                          "         "  O#         T#  $         $  %         F  G         G  )I         .I  lJ         qJ  K         K  L         L  5N         :N  xO         }O  P         P  Q         R  AS         FS  T         T  U         U  
W         W  MX         RX  Y         Y  Z         Z  \         \  Y]         ^]  ^         ^  _         _  "a         'a  eb         jb  c         c  d         d  .f         3f  qg         vg  h         h  i         i  :k         ?k  }l         l  m         m  o         o  Jp         Kp  q         q  r         r  t         t  Vu         Wu  v         v  w         w  y          y  bz         cz  {         {  |         |  +~         ,~  n         o                        7         8  z         {                         C         D             Ɋ         ʊ             O          
            H  5            s          Ē                       ұ             W         X             ݶ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               E!         Z!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 a
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -A-    A-<-x    <-J-|    J-}-\    }-c6-    c6--h    -b-    b-qx-	    qx--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               B!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 8	         9	  u
         v
                        +         ,  h         i                                               I  %          |  X                                              8   !          c!  ;"          "  f#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            B         C                                   N         O                        
         
  Z         [                        #         $  f         g                        /         0  r         s                        ;         <  ~                                 G         H                          "         "  S#         T#  $         $  %          %  &          '  	(          `(  L)          )  *          *  +         F  G         G  -I         .I  pJ         qJ  K         K  L         L  9N         :N  |O         }O  P         P  R         R  ES         FS  T         T  U         U  W         W  QX         RX  Y         Y  Z         Z  \         \  ]]         ^]  ^         ^  _         _  &a         'a  ib         jb  c         c  d         d  2f         3f  ug         vg  h         h  i         i  >k         ?k  l         l  m         m  o         o  Fp         Kp  q         q  r         r  t         t  Ru         Wu  v         v  w         w  y          y  ^z         cz  {         {  |         |  '~         ,~  j         o                        3         8  v         {                        ?         D             Ŋ         ʊ             K         ұ             S         X             ٶ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ?!         N!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 g
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -M-    M-;-x    ;-eJ-|    eJ-}-\    }-c6-    c6--h    -b-    b-kx-	    kx--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               H!         X!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      
           G	         H	  
         
                                   Y         Z                                               .            [  8            e                                                '!  !          M"  %#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ?         C                                   K         O                        
         
  W         [                                  $  c         g                        ,         0  o         s                        8         <  {                                 D         H                          "         "  P#         T#  $         $  %         F  G         G  *I         .I  mJ         qJ  K         K  L         L  6N         :N  yO         }O  P         P  Q         R  BS         FS  T         T  U         U  W         W  NX         RX  Y         Y  Z         Z  \         \  Z]         ^]  ^         ^  _         _  #a         'a  fb         jb  c         c  d         d  /f         3f  rg         vg  h         h  i         i  ;k         ?k  ~l         l  m         m  o         o  Ip         Kp  q         q  r         r  t         t  Uu         Wu  v         v  w         w  y          y  az         cz  {         {  |         |  *~         ,~  m         o                        6         8  y         {                        B         D             Ȋ         ʊ             N          @  -          ~  k                                  8  %         ұ             V         X             ܶ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               E!         Z!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 a
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-<-x    <-J-|    J-}-\    }-c6-    c6--h    -b-    b-qx-	    qx--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               >!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         }                        3	         4	  p
         q
                        &         '  c         d                                               D  !          w  T                                  	            5   !          a!  9"          "  e#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            A         C                        
           M         O                        
         
  Y         [                        "         $  e         g                        .         0  q         s                        :         <  }                                 F         H                          "         "  R#         T#  $         $  %          ),  -          l-  X.          .  /          /  0          51  !2         F  G         G  ,I         .I  oJ         qJ  K         K  L         L  8N         :N  {O         }O  P         P  R         R  DS         FS  T         T  U         U  W         W  PX         RX  Y         Y  Z         Z  \         \  \]         ^]  ^         ^  _         _  %a         'a  hb         jb  c         c  d         d  1f         3f  tg         vg  h         h  i         i  =k         ?k  l         l  m         m  o         o  Gp         Kp  q         q  r         r  t         t  Su         Wu  v         v  w         w  y          y  _z         cz  {         {  |         |  (~         ,~  k         o                        4         8  w         {                        @         D             Ɗ         ʊ  	           L         ұ             T         X             ڶ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               A!         R!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 g
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     ---X     ---     -Y-    Y-6-x    6-UE-|    UE-x-\    x--    --h    -b-    b-kx-	    kx--h
    -K׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               L!         \!  1         
2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         $            H             l  D            h                                    	          M	  8
          
  k                      ,         -  i         j                                    \         ]                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         0            h  T                                             %  2'         7'  u(         z(  )         )  *          +  >,         C,  -         -  .         .  0         0  J1         O1  2         2  3         3  5         5  V6         [6  7         7  8         8  :         $:  b;         g;  <         <  =         =  +?         0?  n@         s@  A         A  B         B  7D         <D  zE         E  F         F   H         H  CI         HI  J         J  K         K  M         M  SN         TN  O         O  P         P  R         R  _S         `S  T         T  U         U  (W         )W  kX         lX  Y         Y  Z         Z  4\         5\  w]         x]  ^         ^  _         _  @a         Aa  b         b  c         c  	e         
e  Lf         Mf  g         g  h         h  j         j  Xk         Yk  l         l  m         m  !o         "o  dp         ep  q         q  r         r  -t         .t  pu         qu  v         v  w         w  9y         :y  |z         }z  {         {  }         }  E~         F~             ˀ         ̀             Q         R             ׅ         ؅             ]         ^                        &         '  i         j                        2         3  u         v                        >         ?                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               @!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 a
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -G-    G-m6-x    m6-E-|    E-Ux-\    Ux--    --h    -b-    b-qx-	    qx--h
    -K׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               D!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                      @            `  8            X                                              '	  
          T
  5                                 7         8  t         u                                   I         J                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                 $            g  ^                     %  6'         7'  y(         z(  )         )  *          +  B,         C,  -         -  .         .  0         0  N1         O1  2         2  3         3  5         5  Z6         [6  7         7  8         8  #:         $:  f;         g;  <         <  =         =  /?         0?  r@         s@  A         A  B         B  ;D         <D  ~E         E  F         F  H         H  GI         HI  J         J  K         K  M         M  ON         TN  O         O  P         P  R         R  [S         `S  T         T  U         U  $W         )W  gX         lX  Y         Y  Z         Z  0\         5\  s]         x]  ^         ^  _         _  <a         Aa  b         b  c         c  e         
e  Hf         Mf  g         g  h         h  j         j  Tk         Yk  l         l  m         m  o         "o  `p         ep  q         q  r         r  )t         .t  lu         qu  v         v  w         w  5y         :y  xz         }z  {         {  |         }  A~         F~             ǀ         ̀  
           M         R             Ӆ         ؅             Y         ^             ߊ           "         '  e         j                        .         3  q         v                        :         ?  }                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               <!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 g
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -S-    S-6-x    6-OE-|    OE-x-\    x--    --h    -b-    b-kx-	    kx--h
    -K׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               J!         X!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         #            F            i  A            d                                              H	  2
          {
  e                      '         (  d         e                                   W         X                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         H  4            l                                  (           %  3'         7'  v(         z(  )         )  *          +  ?,         C,  -         -  .         .  0         0  K1         O1  2         2  3         3  5         5  W6         [6  7         7  8         8   :         $:  c;         g;  <         <  =         =  ,?         0?  o@         s@  A         A  B         B  8D         <D  {E         E  F         F  H         H  DI         HI  J         J  K         K  M         M  RN         TN  O         O  P         P  R         R  ^S         `S  T         T  U         U  'W         )W  jX         lX  Y         Y  Z         Z  3\         5\  v]         x]  ^         ^  _         _  ?a         Aa  b         b  c         c  e         
e  Kf         Mf  g         g  h         h  j         j  Wk         Yk  l         l  m         m   o         "o  cp         ep  q         q  r         r  ,t         .t  ou         qu  v         v  w         w  8y         :y  {z         }z  {         {  }         }  D~         F~             ʀ         ̀             P         R             օ         ؅             \         ^                        %         '  h         j                        1         3  t         v                        =         ?                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               B!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 a
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -;-    ;-s6-x    s6-E-|    E-[x-\    [x--    --h    -b-    b-qx-	    qx--h
    -K׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               @!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         !            B            c  ;            \                                              ,	  
          Y
  ;                                  <         =  y         z                                   N         O                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                     0  '          s  j                                 %  5'         7'  x(         z(  )         )  *          +  A,         C,  -         -  .         .  
0         0  M1         O1  2         2  3         3  5         5  Y6         [6  7         7  8         8  ":         $:  e;         g;  <         <  =         =  .?         0?  q@         s@  A         A  B         B  :D         <D  }E         E  F         F  H         H  FI         HI  J         J  K         K  M         M  PN         TN  O         O  P         P  R         R  \S         `S  T         T  U         U  %W         )W  hX         lX  Y         Y  Z         Z  1\         5\  t]         x]  ^         ^  _         _  =a         Aa  b         b  c         c  e         
e  If         Mf  g         g  h         h  j         j  Uk         Yk  l         l  m         m  o         "o  ap         ep  q         q  r         r  *t         .t  mu         qu  v         v  w         w  6y         :y  yz         }z  {         {  |         }  B~         F~             Ȁ         ̀             N         R             ԅ         ؅             Z         ^                        #         '  f         j                        /         3  r         v                        ;         ?  ~                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               <!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 g
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     ---X     ---     -_-    _-<-x    <-J-|    J-~-\    ~-8-    8--h    -b-    b-߂-	    ߂--h
    -Q׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               S!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      
           G	         H	  
         
                                   Y         Z                                               <            p  H            |                                  8   !          h!  @"          "  p#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            >         C                                   J         O                        
         
  V         [                                 $  b         g                        +         0  n         s                        7         <  z                                  C         H                          "         "  O#         T#  $         $  %         %  '         '  [(         `(  )         )  *         *  $,         ),  g-         l-  .         .  /         /  01         51  s2         x2  3         3  4         4  <6         A6  7         7  8         8  :         
:  H;         M;  <         <  =         =  ?         ?  T@         Y@  A         A  B         B  D         "D  `E         eE  F         F  G         G  )I         .I  lJ         qJ  K          2            f  A            u          Ί              ݌         :  |         }                        E         F             ˚         ̚             Q         R             ן         ؟             ]         ^                        &         '  i         j                        2         3  u         v                        >         ?             Ĳ         Ų             J         K             з         ѷ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               :!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -;-    ;-=<-x    =<-J-|    J-%~-\    %~-8-    8--h    -b-    b-߂-	    ߂--h
    -Q׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ?!         @!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 8	         9	  u
         v
                        +         ,  h         i                                               I  $          |  W                                              J   "!          ~!  V"          "  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            B         C                                   N         O                        
         
  Z         [                        #         $  f         g                        /         0  r         s                        ;         <  ~                                 G         H                          "         "  S#         T#  $         $  %         %  '         '  _(         `(  )         )  *         *  (,         ),  k-         l-  .         .  /         /  41         51  w2         x2  3         3  4         4  @6         A6  7         7  8         8  	:         
:  L;         M;  <         <  =         =  ?         ?  X@         Y@  A         A  B         B  !D         "D  dE         eE  F         F  G         G  -I         .I  pJ         qJ  K          c  d          e  e          @f  g          tg  Lh          h  i         :  x         }                        A         F             ǚ         ̚  
           M         R             ӟ         ؟             Y         ^             ߤ           "         '  e         j                        .         3  q         v                        :         ?  }                    Ų             F         K             ̷         ѷ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               >!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     ---X     ---     -_-    _-<-x    <-J-|    J-~-\    ~-8-    8--h    -b-    b-߂-	    ߂--h
    -Q׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               S!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      
           G	         H	  
         
                                   Y         Z                                               :            m  E            x                                  5   !          f!  >"          "  o#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ?         C                                   K         O                        
         
  W         [                                  $  c         g                        ,         0  o         s                        8         <  {                                 D         H                          "         "  P#         T#  $         $  %         %  '         '  \(         `(  )         )  *         *  %,         ),  h-         l-  .         .  /         /  11         51  t2         x2  3         3  4         4  =6         A6  7         7  8         8  :         
:  I;         M;  <         <  =         =  ?         ?  U@         Y@  A         A  B         B  D         "D  aE         eE  F         F  G         G  *I         .I  mJ         qJ  K          6            j  E            y          Ґ                       :  {         }                        D         F             ʚ         ̚             P         R             ֟         ؟             \         ^                        %         '  h         j                        1         3  t         v                        =         ?             ò         Ų             I         K             Ϸ         ѷ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               :!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -A-    A-1<-x    1<-J-|    J-~-\    ~-8-    8--h    -b-    b-߂-	    ߂--h
    -Q׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               C!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         }                        3	         4	  p
         q
                        &         '  c         d                                               D            w  Q                                              C   !          v!  N"          "  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            A         C                        
           M         O                        
         
  Y         [                        "         $  e         g                        .         0  q         s                        :         <  }                                 F         H                          "         "  R#         T#  $         $  %         %  '         '  ^(         `(  )         )  *         *  ',         ),  j-         l-  .         .  /         /  31         51  v2         x2  3         3  4         4  ?6         A6  7         7  8         8  :         
:  K;         M;  <         <  =         =  ?         ?  W@         Y@  A         A  B         B   D         "D  cE         eE  F         F  G         G  ,I         .I  oJ         qJ  K          i  j          k  k          Dl  m          xm  Pn          n  o         :  y         }                        B         F             Ț         ̚             N         R             ԟ         ؟             Z         ^                        #         '  f         j                        /         3  r         v                        ;         ?  ~                    Ų             G         K             ͷ         ѷ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               <!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     ---X     ---     -Y-    Y-q;-x    q;-J-|    J-Y}-\    Y}-k-    k--h    -ob-    ob-ق-	    ق--h
    -K׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               Q!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             <         =  y         z                        0            |                                                          5            ^  6            _                                                         I          J   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            B         C                                   N         O                        
         
  Z         [                        #         $  f         g                        /         0  r         s                        ;         <  ~                                 G         H                          "         "  S#         T#  $         $  %         %  '         '  [(         `(  )         )  *         *  $,         ),  g-         l-  .         .  /         /  01         51  s2         x2  3         3  4         4  <6         A6  7         7  8         8  :         
:  H;         M;  <         <  =         =  ?         ?  T@         Y@  A         A  B         B  D         "D  `E         eE  F         F  G         G  )I          .I  	J          RJ  -K          vK  QL          L  uM          M  N         zk  l         l  m          n  Bo         Co  p         p  q         q  s         s  Nt         Ot  u         u  v         v  x         x  Zy         [y  z         z  {         {  #}         $}  f~         g~                        /         0  r         s                        ;         <  ~                                 G         H             ͎         Ύ             S         T             ٓ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               1!         2!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     ---X     ---     -_-    _-q;-x    q;-J-|    J-Y}-\    Y}-/-    /--h    -{b-    {b-͂-	    ͂--h
    -K׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               U!         V!   2         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             <         =  y         z                        0                                              +            N  &          q  I            l                                                                     I          J   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            >         C                                   J         O                        
         
  V         [                                 $  b         g                        +         0  n         s                        7         <  z                                  C         H                          "         "  O#         T#  $         $  %         %  '         '  _(         `(  )         )  *         *  (,         ),  k-         l-  .         .  /         /  41         51  w2         x2  3         3  4         4  @6         A6  7         7  8         8  	:         
:  L;         M;  <         <  =         =  ?         ?  X@         Y@  A         A  B         B  !D         "D  dE         eE  F         F  G         G  -I         zk  l         l  m          n  >o         Co  p         p  q         q  s         s  Jt         Ot  u         u  v         v  x         x  Vy         [y  z         z  {         {  }         $}  b~         g~                        +         0  n         s                        7         <  z                                  C         H             Ɏ         Ύ             O         T             Փ          ړ  Ĕ                      F  0          |  f                               >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               5!         :!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -S-    S-};-x    };-J-|    J-e}-\    e}-k-    k--h    -ub-    ub-ق-	    ق--h
    -K׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               M!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             <         =  y         z                        0            |                                                          :            d  <            f                                                         S          T   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            A         C                        
           M         O                        
         
  Y         [                        "         $  e         g                        .         0  q         s                        :         <  }                                 F         H                          "         "  R#         T#  $         $  %         %  '         '  \(         `(  )         )  *         *  %,         ),  h-         l-  .         .  /         /  11         51  t2         x2  3         3  4         4  =6         A6  7         7  8         8  :         
:  I;         M;  <         <  =         =  ?         ?  U@         Y@  A         A  B         B  D         "D  aE         eE  F         F  G         G  *I          N  O          P  P          *Q  R          NR  )S          rS  MT         zk  l         l  m          n  Ao         Co  p         p  q         q  
s         s  Mt         Ot  u         u  v         v  x         x  Yy         [y  z         z  {         {  "}         $}  e~         g~                        .         0  q         s                        :         <  }                                 F         H             ̎         Ύ             R         T             ؓ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               3!         6!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     ---X     ---     -_-    _-};-x    };-J-|    J-e}-\    e}-;-    ;--h    -{b-    {b-͂-	    ͂--h
    -K׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               U!         V!   2         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             <         =  y         z                        0                                  
            .            R  *          v  N            r                                                                     S          T   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ?         C                                   K         O                        
         
  W         [                                  $  c         g                        ,         0  o         s                        8         <  {                                 D         H                          "         "  P#         T#  $         $  %         %  '         '  ^(         `(  )         )  *         *  ',         ),  j-         l-  .         .  /         /  31         51  v2         x2  3         3  4         4  ?6         A6  7         7  8         8  :         
:  K;         M;  <         <  =         =  ?         ?  W@         Y@  A         A  B         B   D         "D  cE         eE  F         F  G         G  ,I         zk  l         l  m          n  ?o         Co  p         p  q         q  s         s  Kt         Ot  u         u  v         v  x         x  Wy         [y  z         z  {         {   }         $}  c~         g~                        ,         0  o         s                        8         <  {                                 D         H             ʎ         Ύ             P         T             ֓            Қ                      T  >            t                               >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               5!         :!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -G-    G-;-x    ;-GJ-|    GJ-}-\    }-/-    /--P    -b-    b-c-	    c-=-D
    =-K׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               H!         L!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      
           G	         H	  
         
                                   Y         Z                                               .            [  ?            l                                               !  !          )"  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            >         C                                   J         O                        
         
  V         [                                 $  b         g                        +         0  n         s                        7         <  z                                  C         H                          "           0  1          T1  ?2         2  3         3  	5         5  L6         Q6  7         7  8         8  :         :  X;         ];  <         <  =         =  !?         &?  d@         i@  A         A  B         B  -D         2D  pE         uE  F         F  G         G  9I         >I  |J         J  K         K  M         M  EN         JN  O         O  P         P  R         R  QS         VS  T         T  U         U  W          W  
X          SX  >Y          Y  rZ         l  n         n  Uo         Zo  p         p  q         q   s         !s  bt         ct  u         u  v         v  *x         +x  my         ny  z         z  {         {  6}         7}  y~         z~                         B         C             Ȅ         Ʉ             N         O             ԉ         Չ             Z         [                        #         v                        >                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               @!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 c
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     ---`     ---     -Y-    Y-;-    ;-J-    J-}-d    }-/-    /-e7-0    e7--    -b-H	    b-c-P
    c-7-
    7-K׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               P!         X!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 8	         9	  u
         v
                        +         ,  h         i                                               I  -          |  `                                              (    !          K!  #"          n"  F#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q   w  operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       \            B         C                                   N         O                        
         
  Z         [                        #         $  f         g                        /         0  r         s                        ;         <  ~                                 G         H                          "         2  3         3  5         5  P6         Q6  7         7  8         8  :         :  \;         ];  <         <  =         =  %?         &?  h@         i@  A         A  B         B  1D         2D  tE         uE  F         F  G         G  =I         >I  J         J  K         K  M         M  IN         JN  O         O  P         P  R         R  US         VS  T         T  U         U  W         l  n         n  Yo         Zo  p         p  q         q  s         !s  ^t         ct  u         u  v         v  &x         +x  iy         ny  z         z  {          {  |          7}  .~          z~  q                                  C             Ą         Ʉ             J         O             Љ         Չ             V         [             ܎                    v                        :                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               D!         X!  1         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 ]
  }                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -A-    A-;-x    ;-GJ-|    GJ-}-\    }-/-    /--P    -b-    b-c-	    c-=-D
    =-K׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               E!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      
           G	         H	  
         
                                   Y         Z                                               .            [  @            m                                               !  !          -"  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            ?         C                                   K         O                        
         
  W         [                                  $  c         g                        ,         0  o         s                        8         <  {                                 D         H                          "         2  3         3  
5         5  M6         Q6  7         7  8         8  :         :  Y;         ];  <         <  =         =  "?         &?  e@         i@  A         A  B         B  .D         2D  qE         uE  F         F  G         G  :I         >I  }J         J  K         K  M         M  FN         JN  O         O  P         P  R         R  RS         VS  T         T  U         U  W          Z  [          [  \          #]  ^          W^  B_          _  v`         l  n         n  Vo         Zo  p         p  q         q  s         !s  at         ct  u         u  v         v  )x         +x  ly         ny  z         z  {         {  5}         7}  x~         z~                         A         C             Ǆ         Ʉ  
           M         O             Ӊ         Չ             Y         [             ߎ           "         v                        =                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               @!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 c
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     ---`     ---     -Y-    Y-;-    ;-J-    J-}-d    }-/-    /-e7-0    e7--    -b-H	    b-c-P
    c-7-
    7-K׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               O!         V!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         }                        3	         4	  p
         q
                        &         '  c         d                                               D  )          w  \                                              %              I!  !"          m"  E#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q   w  operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       \            A         C                        
           M         O                        
         
  Y         [                        "         $  e         g                        .         0  q         s                        :         <  }                                 F         H                          "         2  3         3  5         5  O6         Q6  7         7  8         8  :         :  [;         ];  <         <  =         =  $?         &?  g@         i@  A         A  B         B  0D         2D  sE         uE  F         F  G         G  <I         >I  J         J  K         K  M         M  HN         JN  O         O  P         P  R         R  TS         VS  T         T  U         U  W         l  n         n  Xo         Zo  p         p  q         q  s         !s  _t         ct  u         u  v         v  'x         +x  jy         ny  z         z  {         {  3}         7}  v~         z~                         ?          C  :            }          Ʉ                        O  F           щ         Չ             W         [             ݎ                     v                        ;                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               B!         T!  1         1  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 ]
  }                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     -5-    5-k;-    k;-J-    J-S}-l    S}--    --P    --    -b-h	    b-u-p
    u--
    -簳-    簳-E׳-L                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               C!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             <         =  y         z                        0            |                                                          5            ^  6            _                                                         I          J   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        %  F&         K&  '         '  (         (  *         *  R+         W+  ,         ,  -         -  /          /  ^0         c0  1         1  2         2  '4         ,4  j5         o5  6         6  7         7  39         89  v:         {:  ;         ;  <         =  ?>          D>  ;?          ?  ~@          @  A          B  C          PC  GD         D  E         E  G         G  WH         \H  I         I  J         J   L         %L  cM         hM  N         N  O         O  ,Q         1Q  oR         tR  S         S  T         T  8V         =V  {W         W  X         X  Z         Z  D[         I[  \         \  ]         ]  _         _  P`         U`  a         a  b         b  d         d  \e         ae  f         f  g         g  %i         *i  hj         mj  k         k  l         l  5n         6n  xo         yo  p         p  q         q  As         Bs  t         t  u         u  
w         w  Mx         Nx  y         y  z         z  |         |  Y}         Z}  ~         ~             "         #  e         f                        .         /  q         r                        :         ;  }         ~                        F         G             ̒         ͒             N         S                                b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               K!         ^!  1         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
  	         
                                b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     2
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     ---`     ---     -_-    _-k;-    k;-J-    J-S}-d    S}--    --p    -b-    b-u-
    u--
    -۰-t    ۰-E׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               R!         `!  2         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             <         =  y         z                        0                                              +            N  &          q  I            l                                                                     I          J   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                 (            d  W                     %  J&         K&  '         '  (         (  *         *  V+         W+  ,         ,  -         -  /          /  b0         c0  1         1  2         2  +4         ,4  n5         o5  6         6  7         7  79         89  z:         {:  ;         ;   =         =  C>         D>  ?         ?  @         @  B         B  OC         PC  D         D  E         E  G         G  [H         \H  I         I  J         J  $L         %L  gM         hM  N         N  O         O  0Q         1Q  sR         tR  S         S  T         T  <V         =V  W         W  X         X  Z         Z  H[         I[  \         \  ]         ]  _         _  T`         U`  a         a  b         b  d         d  `e         ae  f         f  g         g  )i         *i  lj         mj  k         k  l         l  1n         6n  to         yo  p         p  q         q  =s         Bs  t         t  u         u  w         w  Ix         Nx  y         y  z         z  |         |  U}         Z}  ~         ~                      #  a         f                        *         /  m         r                        6         ;  y         ~                        B         G             Ȓ         ͒             R         S                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               S!         j!  2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
  	         
        
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     &
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     -5-    5-q;-    q;-J-    J-Y}-l    Y}--    -s-P    s--    -b-h	    b-u-p
    u--
    -簳-    簳-E׳-L                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               A!         F!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             <         =  y         z                        0            |                                                          :            d  <            f                                                         S          T   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       h        %  G&         K&  '         '  (         (  *         *  S+         W+  ,         ,  -         -  /          /  _0         c0  1         1  2         2  (4         ,4  k5         o5  6         6  7         7  49         89  w:         {:  ;         ;  <         =  @>         D>  ?         ?  @         @  	B         B  LC         PC  D          D  E          E  F         G  XH         \H  I         I  J         J  !L         %L  dM         hM  N         N  O         O  -Q         1Q  pR         tR  S         S  T         T  9V         =V  |W         W  X         X  Z         Z  E[         I[  \         \  ]         ]  _         _  Q`         U`  a         a  b         b  d         d  ]e         ae  f         f  g         g  &i         *i  ij         mj  k         k  l         l  4n         6n  wo         yo  p         p  q         q  @s         Bs  t         t  u         u  	w         w  Lx         Nx  y         y  z         z  |         |  X}         Z}  ~         ~             !         #  d         f                        -         /  p         r                        9         ;  |         ~                        E         G             ˒          ͒  ē                      S  J                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               I!         ^!  1         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              
                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
  	         
                                b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     1
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     ---`     ---     -Y-    Y-q;-    q;-J-    J-Y}-d    Y}--    --p    -b-    b-u-
    u--
    -۰-t    ۰-E׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               P!         \!  1         
2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             <         =  y         z                        0                                  
            .            R  *          v  N            r                                                                     S          T   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                        !          T!  G"          "  #          #  $         %  I&         K&  '         '  (         (  *         *  U+         W+  ,         ,  -         -  /          /  a0         c0  1         1  2         2  *4         ,4  m5         o5  6         6  7         7  69         89  y:         {:  ;         ;  <         =  B>         D>  ?         ?  @         @  B         B  NC         PC  D         D  E         E  G         G  ZH         \H  I         I  J         J  #L         %L  fM         hM  N         N  O         O  /Q         1Q  rR         tR  S         S  T         T  ;V         =V  ~W         W  X         X  Z         Z  G[         I[  \         \  ]         ]  _         _  S`         U`  a         a  b         b  d         d  _e         ae  f         f  g         g  (i         *i  kj         mj  k         k  l         l  2n         6n  uo         yo  p         p  q         q  >s         Bs  t         t  u         u  w         w  Jx         Nx  y         y  z         z  |         |  V}         Z}  ~         ~                      #  b         f                        +         /  n         r                        7         ;  z         ~                         C         G             ɒ         ͒             Q         S                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               S!         j!  2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
  	         
        
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     %
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     -5-    5-6-    6-gE-    gE-x-l    x-{9-    {9--x    -b-	    b-u-
    u--
    --|    -%ֳ-    %ֳ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               A!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         )            R  *          {  S            |                                   3  	          f	  D
          
  w                    	  E         F                                   8         9  u         v                        ,                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |        D%  &          &  g'          '  (          (  )          *  *          7+  ,         I  J         J  4L         5L  wM         xM  N         N  O         O  @Q         AQ  R         R  S         S  	U         
U  LV         MV  W         W  X         X  Z         Z  X[         Y[  \         \  ]         ]  !_         "_  d`         e`  a         a  b         b  -d         .d  pe         qe  f         f  g         g  9i         :i  |j         }j  k         k  m         m  An         Fn  o         o  p         p  r         r  Qs         Rs  t         t  u         u  w         w  ]x         ^x  y         y  z         z  &|         '|  i}         j}  ~         ~             2         3  u         v                        >         ?             Ĉ         ň             J         K             Ѝ         э             V         W             ܒ         ݒ              ^         c                        +         ,  j         o                        3         8  v         {                        ?         D             ť         ʥ             K         P             Ѫ         ֪             W         \             ݯ                     %  c         h                        ,         1  o         t                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               @!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
  	         
        
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
  	                              b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            0     -,-X     ,--     -M-    M-s6-x    s6-E-|    E-[x-\    [x-w-    w--h    -b-    b-n-	    n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               J!         X!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         #            F            i  A            d                                  	            6	  
          c
  A                      	         
  F         G                                              X         Y                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            =         >  {         |                        5         6  s         t             	         	  -         .  k         l                        %         &  c         d                                   [         \                                   S         T                                   K          L   !         !  "         "  $         $  C%          %  t&          &  '          '  (          )  )          ,*  +         G  H         H  0J         2J  nK         pK  L         L  M         M  (O         *O  fP         hP  Q         Q  R         R   T         "T  ^U         `U  V         V  W         W  Y         Y  VZ         XZ  [         [  \         \  ^         ^  N_         P_  `         `  a         a  c         
c  Fd         Hd  e         e  f         f   h         h  >i         @i  |j         ~j  k         k  l         n  o         o  *q         ,q  hr         jr  s         s  t         t  "v         $v  `w         bw  x         x  y         y  {         {  X|         Z|  }         }  ~         ~             P         R             ̃         ΃  
           H         J             Ĉ         ƈ             @         B  ~                                 8         :  v         x                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               H!         X!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     -
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     -5-    5-6-    6-mE-    mE-x-l    x-{9-    {9--x    -b-	    b-u-
    u--
    --|    -%ֳ-    %ֳ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               ?!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         *            T  ,          ~  V                                              8  	          k	  J
          
  }                      J         K                                   =         >  z         {                        1                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |        D%  &          c,  C-          -  o.          .  /          /  0          1  1         I  J         J  3L         5L  vM         xM  N         N  O         O  ?Q         AQ  R         R  S         S  U         
U  KV         MV  W         W  X         X  Z         Z  W[         Y[  \         \  ]         ]   _         "_  c`         e`  a         a  b         b  ,d         .d  oe         qe  f         f  g         g  8i         :i  {j         }j  k         k  m         m  Bn         Fn  o         o  p         p  r         r  Ps         Rs  t         t  u         u  w         w  \x         ^x  y         y  z         z  %|         '|  h}         j}  ~         ~             1         3  t         v                        =         ?             È         ň             I         K             ύ         э             U         W             ے         ݒ              _         c                        *         ,  k         o                        4         8  w         {                        @         D             ƥ         ʥ  	           L         P             Ҫ         ֪             X         \             ޯ           !         %  d         h                        -         1  p         t                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               <!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
  	         
        
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
  	                              b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            0     -,-X     ,--     -G-    G-y6-x    y6-E-|    E-ax-\    ax-w-    w--h    -b-    b-n-	    n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               H!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         $            H             l  D            h                                              ;	  
          h
  G                                 K         L                                             !  ]         ^                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            <         >  z         |                        4         6  r         t             	         	  ,         .  j         l                        $         &  b         d                                   Z         \                                   R         T                                   J          L   !         !  "         "  $         $  B%          P+  (,          t,  L-          -  p.          .  /          /  0         G  H         H  1J         2J  oK         pK  L         L  M         M  )O         *O  gP         hP  Q         Q  R         R  !T         "T  _U         `U  V         V  W         W  Y         Y  WZ         XZ  [         [  \         \  ^         ^  O_         P_  `         `  a         a  	c         
c  Gd         Hd  e         e  f         f  h         h  ?i         @i  }j         ~j  k         k  l         n  o         o  +q         ,q  ir         jr  s         s  t         t  #v         $v  aw         bw  x         x  y         y  {         {  Y|         Z|  }         }  ~         ~             Q         R             ̓         ΃             I         J             ň         ƈ             A         B                                   9         :  w         x                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               F!         T!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     .
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-_;-x    _;-I-|    I-G}-\    G}-7-    7--h    -b-    b-	n-	    	n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               @!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             <         =  y         z                        0            |                                                          5            ^  6            _                                                         I          J   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            >         @  ~                                  >         @  ~                      	          
  >         @  ~                                  >         @  ~                                  >         @  ~                                  >         @  ~                                  >         @  ~             !         !  "          #  >$         @$  ~%          2  2          /3  4          Y4  15          5  [6          6  7         K  L         L  /N         0N  oO         pO  P         P  Q         Q  /S         0S  oT         pT  U         U  V         V  /X         0X  oY         pY  Z         Z  [         [  /]         0]  o^         p^  _         _  `         `  /b         0b  oc         pc  d         d  e         e  /g         0g  oh         ph  i         i  j         j  /l         0l  om         pm  n         n  o         o  /q         0q  or           2         4  r         t                        2         4  r         t                        2         4  r         t                        2         4  r         t                        2         4  r         t                        2         4  r         t                        2         4  r         t                        2                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               C!         V!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     +
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -A-    A-;-x    ;-}J-|    }J-}-\    }--    --h    -b-    b-ex-	    ex--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               D!         X!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 8	         9	  u
         v
                        +         ,  h         i                                               I  -          |  `                                              (    !          K!  #"          n"  F#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ?         @                                    ?         @                        	          
  ?         @                                    ?         @                                    ?         @                                    ?         @                                    ?         @               !         !  "          #  ?$         @$  %         %  &         &  '          (  ?)         @)  *         *  +         +  ,          -  ?.         @.  /         /  0         0  1          2  ?3         @3  4         4  5         5  6          7  ?8         @8  9         9  :         :  ;          <  ?=         @=  >         >  ?         ?  @          A  ?B         @B  C         C  D         D  E          F  ?G         @G  H         H  I         I  J          K  >L         @L  ~M         M  N         N  O          P  >Q         @Q  ~R         R  S         S  T          U  >V         @V  ~W         W  X         X  Y          Z  >[         @[  ~\         \  ]         ]  ^          _  >`         @`  ~a         a  b         b  c          d  >e         @e  ~f         f  g         g  h          i  >j         @j  ~k         k  l         l  m          z  {          '|  }          j}  ^~          ~                          ^         `                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               O!         ^!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 g
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-k;-x    k;-J-|    J-S}-\    S}-7-    7--h    -b-    b-	n-	    	n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               @!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             <         =  y         z                        0            |                                                          :            d  <            f                                                         S          T   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ?         @                                    ?         @                        	          
  ?         @                                    ?         @                                    ?         @                                    ?         @                                    ?         @               !         !  "          #  ?$         @$  %          7  8          9  9          +:  ;          U;  -<          <  W=         K  L         L  .N         0N  nO         pO  P         P  Q         Q  .S         0S  nT         pT  U         U  V         V  .X         0X  nY         pY  Z         Z  [         [  .]         0]  n^         p^  _         _  `         `  .b         0b  nc         pc  d         d  e         e  .g         0g  nh         ph  i         i  j         j  .l         0l  nm         pm  n         n  o         o  .q         0q  nr           3         4  s         t                        3         4  s         t                        3         4  s         t                        3         4  s         t                        3         4  s         t                        3         4  s         t                        3         4  s         t                        3                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               C!         V!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     *
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -;-    ;-;-x    ;-}J-|    }J-}-\    }-#-    #--h    -b-    b-ex-	    ex--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               B!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         }                        3	         4	  p
         q
                        &         '  c         d                                               D  )          w  \                                              %              I!  !"          m"  E#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            >         @  ~                                  >         @  ~                      	          
  >         @  ~                                  >         @  ~                                  >         @  ~                                  >         @  ~                                  >         @  ~             !         !  "          #  >$         @$  ~%         %  &         &  '          (  >)         @)  ~*         *  +         +  ,          -  >.         @.  ~/         /  0         0  1          2  >3         @3  ~4         4  5         5  6          7  >8         @8  ~9         9  :         :  ;          <  >=         @=  ~>         >  ?         ?  @          A  >B         @B  ~C         C  D         D  E          F  >G         @G  ~H         H  I         I  J          K  ?L         @L  M         M  N         N  O          P  ?Q         @Q  R         R  S         S  T          U  ?V         @V  W         W  X         X  Y          Z  ?[         @[  \         \  ]         ]  ^          _  ?`         @`  a         a  b         b  c          d  ?e         @e  f         f  g         g  h          i  ?j         @j  k         k  l         l  m          3  '          v  j                                  ?  3            _         `                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               M!         ^!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 g
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-6-x    6-aE-|    aE-x-\    x-'0-    '0--h    -b-    b-	n-	    	n--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               @!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         )            R  *          {  S            |                                   3  	          f	  D
          
  w                    	  E         F                                   8         9  u         v                        ,                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            <         >  z         |                        4         6  r         t             	         	  ,         .  j         l                        $         &  b         d                                   Z         \                                   R         T                                   J          L   !         !  "         "  $         $  B%         H  !J         "J  _K         `K  L         L  M         M  O         O  WP         XP  Q         Q  R         R  T         T  OU         PU  V         V  W         W  	Y         
Y  GZ         HZ  [         [  \         \  ^         ^  ?_         @_  }`         ~`  a         a  b         b  7d         8d  ue         ve  f         f  g         g  /i         0i  mj         nj  k         k  l         l  'n         o  q         q  Er         Fr  s         s  t         t  u          v  =w         >w  {x         |x  y         y  z         z  5|         6|  s}         t}  ~         ~             -         .  k         l                        %         &  c         d             ߉                      [         \             ׎         ؎             S         T             ϓ         Г            ,            V  .            X                      ԯ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               B!         T!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     +
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -;-    ;-m6-x    m6-E-|    E-Ux-\    Ux-y-    y--h    -b-    b-ex-	    ex--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               D!         X!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         #            F            i  A            d                                  	            6	  
          c
  A                      	         
  F         G                                              X         Y                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                 (            d  T                     p&  '         '  (         (  8*         9*  {+         |+  ,         ,  .         .  D/         E/  0         0  1         1  3         3  P4         Q4  5         5  6         6  8         8  \9         ]9  :         :  ;         ;  %=         &=  h>         i>  ?         ?  @         @  1B         2B  tC         uC  D         D  E         E  =G         >G  H         H  I         I  K         K  IL         JL  M         M  N         N  P         P  QQ         VQ  R         R  S         S  U         U  ]V         bV  W         W  X         X  &Z         +Z  i[         n[  \         \  ]         ]  2_         7_  u`         z`  a         a  b          c  >d         Cd  e         e  f         f  h         h  Ji         Oi  j         j  k         k  m         m  Vn         [n  o         o  p         p  r         $r  bs         gs  t         t  u         u  +w         0w  nx         sx  y         y  z         z  7|         <|  z}         }  ~         ~              C         H             Ƀ         ΃             O         T             Ո         ڈ             [         `                        $         )  g         l                        0         5  s         x                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               F!         T!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 g
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-6-x    6-gE-|    gE-x-\    x-5-    5--h    -b-    b-	n-	    	n--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               @!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         *            T  ,          ~  V                                              8  	          k	  J
          
  }                      J         K                                   =         >  z         {                        1                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            =         >  {         |                        5         6  s         t             	         	  -         .  k         l                        %         &  c         d                                   [         \                                   S         T                                   K          L   !         !  "         "  $         $  C%         H   J         "J  ^K         `K  L         L  M         M  O         O  VP         XP  Q         Q  R         R  T         T  NU         PU  V         V  W         W  Y         
Y  FZ         HZ  [         [  \         \   ^         ^  >_         @_  |`         ~`  a         a  b         b  6d         8d  te         ve  f         f  g         g  .i         0i  lj         nj  k         k  l         l  &n         o  q         q  Dr         Fr  s         s  t         t  u          v  <w         >w  zx         |x  y         y  z         z  4|         6|  r}         t}  ~         ~             ,         .  j         l                        $         &  b         d             މ                      Z         \             ֎         ؎             R         T             Γ         Г              ֱ          (             R  *          |  T            ~                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               @!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     *
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-s6-x    s6-E-|    E-[x-\    [x-y-    y--h    -b-    b-ex-	    ex--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               B!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         $            H             l  D            h                                              ;	  
          h
  G                                 K         L                                             !  ]         ^                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                        !          T!  D"          "  #          #  $         p&  '         '  (         (  7*         9*  z+         |+  ,         ,   .         .  C/         E/  0         0  1         1  3         3  O4         Q4  5         5  6         6  8         8  [9         ]9  :         :  ;         ;  $=         &=  g>         i>  ?         ?  @         @  0B         2B  sC         uC  D         D  E         E  <G         >G  H         H  I         I  K         K  HL         JL  M         M  N         N  P         P  RQ         VQ  R         R  S         S  U         U  ^V         bV  W         W  X         X  'Z         +Z  j[         n[  \         \  ]         ]  3_         7_  v`         z`  a         a  b          c  ?d         Cd  e         e  f         f  h         h  Ki         Oi  j         j  k         k  m         m  Wn         [n  o         o  p         p   r         $r  cs         gs  t         t  u         u  ,w         0w  ox         sx  y         y  z         z  8|         <|  {}         }  ~         ~             D         H             ʃ         ΃             P         T             ֈ         ڈ             \         `                        %         )  h         l                        1         5  t         x                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               B!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 g
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -;-    ;-;-x    ;-/J-|    /J-}}-\    }}-7-    7--h    -]b-    ]b--	    --h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               D!         X!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      
           G	         H	  
         
                                   Y         Z                                               .            [  ?            l                                               !  !          )"  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            >         C                                   J         O                        
         
  V         [                                 $  b         g                        +         0  n         s                        7         <  z                                  C         H                          "         "  O#         T#  $         $  %         %  '         '  [(         `(  )         )  *         *  $,         ),  g-         l-  .         .  /         /  01         51  s2         x2  3         3  4         4  <6         A6  7         7  8         8  :         
:  H;         M;  <         <  =         =  ?         ?  T@         Y@  A         A  B         B  D         "D  `E         eE  F         F  G         G  )I         .I  lJ         qJ  K         n  o         o  %q         &q  hr          ir  Ss          s  t          t  u          u  v          )w  x         	  K         L             љ         ҙ             W         X             ݞ         ޞ            !  c         d                        ,         -  o         p                        8         9  {         |                        D         E             ʱ         ˱             P         Q             ֶ         ׶                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               5!         6!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-;-x    ;-wJ-|    wJ-}-\    }-7-    7--h    -b-    b--	    --h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               @!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 8	         9	  u
         v
                        +         ,  h         i                                               I  -          |  `                                              (    !          K!  #"          n"  F#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            B         C                                   N         O                        
         
  Z         [                        #         $  f         g                        /         0  r         s                        ;         <  ~                                 G         H                          "         "  S#         T#  $         $  %         %  '         '  _(         `(  )         )  *         *  (,         ),  k-         l-  .         .  /         /  41         51  w2         x2  3         3  4         4  @6         A6  7         7  8         8  	:         
:  L;         M;  <         <  =         =  ?         ?  X@         Y@  A         A  B         B  !D         "D  dE         eE  F         F  G         G  -I         .I  pJ         qJ  K          L  vM          M  N          N  O          
P  P          .Q  R         n  o         o  !q         &q  dr         	  G         L             ͙         ҙ             S         X             ٞ         ޞ           !  _         d                        (         -  k         p                        4         9  w         |                        @         E             Ʊ         ˱  	           L         Q             Ҷ         ׶                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               G!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-;-x    ;-5J-|    5J-}-\    }-7-    7--h    -cb-    cb--	    --h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               B!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                      
           G	         H	  
         
                                   Y         Z                                               .            [  @            m                                               !  !          -"  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ?         C                                   K         O                        
         
  W         [                                  $  c         g                        ,         0  o         s                        8         <  {                                 D         H                          "         "  P#         T#  $         $  %         %  '         '  \(         `(  )         )  *         *  %,         ),  h-         l-  .         .  /         /  11         51  t2         x2  3         3  4         4  =6         A6  7         7  8         8  :         
:  I;         M;  <         <  =         =  ?         ?  U@         Y@  A         A  B         B  D         "D  aE         eE  F         F  G         G  *I         .I  mJ         qJ  K         n  o         o  $q         &q  gr          Yx  Cy          y  sz          z  {          {  |          }  ~         	  J         L             Й         ҙ             V         X             ܞ         ޞ           !  b         d                        +         -  n         p                        7         9  z         |                         C         E             ɱ         ˱             O         Q             ն         ׶                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               9!         :!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-;-x    ;-wJ-|    wJ-}-\    }-7-    7--h    -b-    b--	    --h
    -3׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               @!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         }                        3	         4	  p
         q
                        &         '  c         d                                               D  )          w  \                                              %              I!  !"          m"  E#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            A         C                        
           M         O                        
         
  Y         [                        "         $  e         g                        .         0  q         s                        :         <  }                                 F         H                          "         "  R#         T#  $         $  %         %  '         '  ^(         `(  )         )  *         *  ',         ),  j-         l-  .         .  /         /  31         51  v2         x2  3         3  4         4  ?6         A6  7         7  8         8  :         
:  K;         M;  <         <  =         =  ?         ?  W@         Y@  A         A  B         B   D         "D  cE         eE  F         F  G         G  ,I         .I  oJ         qJ  K          RR  *S          vS  NT          T  rU          U  V          V  W         n  o         o  "q         &q  er         	  H         L             Ι         ҙ             T         X             ڞ         ޞ           !  `         d                        )         -  l         p                        5         9  x         |                        A         E             Ǳ         ˱  
           M         Q             Ӷ         ׶                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               G!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,-7-     7--    -6-x    6-7E-|    7E-x-\    x--    --h    -{b-    {b-Sx-	    Sx--h
    -+-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     H                       P!  1          1  p=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         #            F            i  A            d                                              B	  $
          s
  U                      !         "  `         a                                   Q         R                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                                 $            g  [                     &  '         '  )         ")  `*         e*  +         +  ,         ,  ).         ..  l/         q/  0         0  1         1  53         :3  x4         }4  5         5  6         7  A8         F8  9         9  :         :  
<         <  M=         R=  >         >  ?         ?  A         A  YB         ^B  C         C  D         D  "F         'F  eG         jG  H         H  I         I  .K         3K  qL         vL  M         M  N         N  >P         ?P  Q         Q  R         R  T         T  JU         KU  V         V  W         W  Y         Y  VZ         WZ  [         [  \         \  ^          ^  b_         c_  `         `  a         a  +c         ,c  nd         od  e         e  f         f  7h         8h  zi         {i  j         j   l         l  Cm         Dm  n         n  o         o  q         q  Or         Pr  s         s  t         t  v         v  [w         \w  x         x  y         y  ${         %{  g|         h|  }         }  ~         ~  0         1  s         t                        <         =                      Ç             H         I             Ό         ό             T         U             ڑ         ۑ             `         a                        )                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               G!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 c
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,-7-     7--    -m6-x    m6-E-|    E-Ux-\    Ux--    --h    -cb-    cb-Yx-	    Yx--h
    -+-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     L                       L!  1          1  n=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     :            W  /          t  L            w                                  	  
          M
  3         |                        8         9  w         x                                    ^         _                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         <  6            y                                  H  B         &  '         '  !)         ")  d*         e*  +         +  ,         ,  -.         ..  p/         q/  0         0  1         1  93         :3  |4         }4  5         5  7         7  E8         F8  9         9  :         :  <         <  Q=         R=  >         >  ?         ?  A         A  ]B         ^B  C         C  D         D  &F         'F  iG         jG  H         H  I         I  2K         3K  uL         vL  M         M  N         N  :P         ?P  }Q         Q  R         R  T         T  FU         KU  V         V  W         W  Y         Y  RZ         WZ  [         [  \         \  ^          ^  ^_         c_  `         `  a         a  'c         ,c  jd         od  e         e  f         f  3h         8h  vi         {i  j         j  k         l  ?m         Dm  n         n  o         o  q         q  Kr         Pr  s         s  t         t  v         v  Ww         \w  x         x  y         y   {         %{  c|         h|  }         }  ~         ~  ,         1  o         t                        8         =  {                    Ç             D         I             ʌ         ό             P         U             ֑         ۑ             \         a                        %                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ;!         <!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 e
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -{,-X     {,-1-     1--    -6-x    6-=E-|    =E-x-\    x--    --h    -{b-    {b-Sx-	    Sx--h
    -+-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     G                       L!  1          1  i=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         $            H             l  D            h                                              G	  *
          x
  [                      &         '  e         f                                   V         W                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                     0  $          s  g                                 &  '         '  )         ")  a*         e*  +         +  ,         ,  *.         ..  m/         q/  0         0  1         1  63         :3  y4         }4  5         5  6         7  B8         F8  9         9  :         :  <         <  N=         R=  >         >  ?         ?  A         A  ZB         ^B  C         C  D         D  #F         'F  fG         jG  H         H  I         I  /K         3K  rL         vL  M         M  N         N  =P         ?P  Q         Q  R         R  T         T  IU         KU  V         V  W         W  Y         Y  UZ         WZ  [         [  \         \  ^          ^  a_         c_  `         `  a         a  *c         ,c  md         od  e         e  f         f  6h         8h  yi         {i  j         j  k         l  Bm         Dm  n         n  o         o  q         q  Nr         Pr  s         s  t         t  v         v  Zw         \w  x         x  y         y  #{         %{  f|         h|  }         }  ~         ~  /         1  r         t                        ;         =  ~                    Ç             G         I             ͌         ό             S         U             ّ         ۑ             _         a                        (                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               G!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 c
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -{,-X     {,-1-     1--    -m6-x    m6-E-|    E-Ux-\    Ux--    --h    -ib-    ib-Yx-	    Yx--h
    -+-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     K                       H!  1          1  g=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     <            Z  2          x  P            }                                  #	  

          R
  9                                 =         >  |         }                        $         %  c         d                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                         !          "  #          T#  N$          $  %         &  '         '   )         ")  c*         e*  +         +  ,         ,  ,.         ..  o/         q/  0         0  1         1  83         :3  {4         }4  5         5  7         7  D8         F8  9         9  :         :  <         <  P=         R=  >         >  ?         ?  A         A  \B         ^B  C         C  D         D  %F         'F  hG         jG  H         H  I         I  1K         3K  tL         vL  M         M  N         N  ;P         ?P  ~Q         Q  R         R  T         T  GU         KU  V         V  W         W  Y         Y  SZ         WZ  [         [  \         \  ^          ^  __         c_  `         `  a         a  (c         ,c  kd         od  e         e  f         f  4h         8h  wi         {i  j         j  k         l  @m         Dm  n         n  o         o  	q         q  Lr         Pr  s         s  t         t  v         v  Xw         \w  x         x  y         y  !{         %{  d|         h|  }         }  ~         ~  -         1  p         t                        9         =  |                    Ç             E         I             ˌ         ό             Q         U             ב         ۑ             ]         a                        &                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ?!         @!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 e
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -i,-X     i,--     --    -;-x    ;-qJ-|    qJ-}-\    }-9-    9--h    -b-    b--	    --h
    -1-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     D                       @!  1          1  T=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 C	         D	  
         
                        *         +  i         j                                               F  "          u  Q                                              !              H!   "          o"  G#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            >         C                                   J         O                        
         
  V         [                                 $  b         g                        +         0  n         s                        7         <  z                                  C         H                          "         "  O#         T#  $         $  %         %  '         '  [(         `(  )         )  *         *  $,         ),  g-         l-  .         .  /         /  01         51  s2         x2  3         3  4         4  <6         A6  7         7  8         8  :         
:  H;         M;  <         <  =         =  ?         ?  T@         Y@  A         A  B         B  D         "D  `E         eE  F         F  G         G  )I         .I  lJ         qJ  K          c  d          d  e          
f  f          2g  
h          Zh  2i         Z             ߖ           "         #  e         f                        .         /  q         r                        :         ;  }         ~                        F         G             ̩         ͩ             R         S             خ         ٮ             ^         _                        '         (  j         k                        3                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               F!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -Q,-X     Q,--     --    -;-x    ;-J-|    J-}-\    }-9-    9--h    -ub-    ub--	    --h
    -1-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     A                       0!  q1          1  9=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         }                        9	         :	  x
         y
                        *         +  i         j                                               J  "          {  S                                  
            7   !          d!  <"          "  i#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            B         C                                   N         O                        
         
  Z         [                        #         $  f         g                        /         0  r         s                        ;         <  ~                                 G         H                          "         "  S#         T#  $         $  %         %  '         '  _(         `(  )         )  *         *  (,         ),  k-         l-  .         .  /         /  41         51  w2         x2  3         3  4         4  @6         A6  7         7  8         8  	:         
:  L;         M;  <         <  =         =  ?         ?  X@         Y@  A         A  B         B  !D         "D  dE         eE  F         F  G         G  -I         .I  pJ         qJ  K          K  L          L  M          N  N          >O  P          lP  DQ         Z             ۖ                    #  a         f                        *         /  m         r                        6         ;  y         ~                        B         G             ȩ         ͩ             N         S             Ԯ         ٮ             Z         _                        #         (  f         k                        /                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               >!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -o,-X     o,-%-     %--    -;-x    ;-wJ-|    wJ-}-\    }-9-    9--h    -{b-    {b--	    --h
    -1-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     E                       D!  1          1  [=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 C	         D	  
         
                        *         +  i         j                                               F  #          u  R                                              #              K!  #"          s"  K#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ?         C                                   K         O                        
         
  W         [                                  $  c         g                        ,         0  o         s                        8         <  {                                 D         H                          "         "  P#         T#  $         $  %         %  '         '  \(         `(  )         )  *         *  %,         ),  h-         l-  .         .  /         /  11         51  t2         x2  3         3  4         4  =6         A6  7         7  8         8  :         
:  I;         M;  <         <  =         =  ?         ?  U@         Y@  A         A  B         B  D         "D  aE         eE  F         F  G         G  *I         .I  mJ         qJ  K          i  Zj          j  k          k  l          l  m          "n  n         Z             ޖ           !         #  d         f                        -         /  p         r                        9         ;  |         ~                        E         G             ˩         ͩ             Q         S             ׮         ٮ             ]         _                        &         (  i         k                        2                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               B!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -K,-X     K,--     --    -<-x    <-J-|    J-}-\    }-9-    9--h    -ob-    ob--	    --h
    -1-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                     @                       .!  n1          1  5=                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         }                        9	         :	  x
         y
                        *         +  i         j                                               J  #          {  T                                              9   !          g!  ?"          "  m#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            A         C                        
           M         O                        
         
  Y         [                        "         $  e         g                        .         0  q         s                        :         <  }                                 F         H                          "         "  R#         T#  $         $  %         %  '         '  ^(         `(  )         )  *         *  ',         ),  j-         l-  .         .  /         /  31         51  v2         x2  3         3  4         4  ?6         A6  7         7  8         8  :         
:  K;         M;  <         <  =         =  ?         ?  W@         Y@  A         A  B         B   D         "D  cE         eE  F         F  G         G  ,I         .I  oJ         qJ  K          Q  rR          R  S          S  T          $U  U          RV  *W         Z             ܖ                    #  b         f                        +         /  n         r                        7         ;  z         ~                         C         G             ɩ         ͩ             O         S             ծ         ٮ             [         _                        $         (  g         k                        0                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               <!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
            0     -,-X     ,--     -;-    ;-;-x    ;-J-|    J-k}-\    k}--    -q-h    q-b-    b--	    --h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               E!         V!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             >         ?  }         ~                        :                                              -            T  ,          {  S            v                                                          .         /  m          n   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            B         C                                   N         O                        
         
  Z         [                        #         $  f         g                        /         0  r         s                        ;         <  ~                                 G         H                          "         "  S#         T#  $         $  %         %  '         '  [(         `(  )         )  *         *  $,         ),  g-         l-  .         .  /         /  01         51  s2         x2  3         3  4         4  <6         A6  7         7  8         8  :         
:  H;         M;  <         <  =         =  ?         ?  T@         Y@  A         A  B         B  D         "D  `E         eE  F         F  G         G  )I          I  J          K  K          DL  +M          vM  ]N          N  O         l  m         m  o         o  Rp         Sp  q         q  r         r  t         t  ^u         _u  v         v  w         w  'y         (y  jz         kz  {         {  |         |  3~         4~  v         w                        ?         @             Ņ         ƅ           	  K         L             ъ         Ҋ             W         X             ݏ         ޏ            !  c           ؔ         ٔ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     C                       :!  }1          1  J=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -/-    /-;-x    ;-J-|    J-k}-\    k}--    --h    --b-    -b--	    --h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               A!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             >         ?  }         ~                        :                                              I  !          v  N            {                                                                      .         /  m          n   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            >         C                                   J         O                        
         
  V         [                                 $  b         g                        +         0  n         s                        7         <  z                                  C         H                          "         "  O#         T#  $         $  %         %  '         '  _(         `(  )         )  *         *  (,         ),  k-         l-  .         .  /         /  41         51  w2         x2  3         3  4         4  @6         A6  7         7  8         8  	:         
:  L;         M;  <         <  =         =  ?         ?  X@         Y@  A         A  B         B  !D         "D  dE         eE  F         F  G         G  -I          U  V          W  W          8X  Y          jY  NZ          Z  [         l  m         m  o         o  Np         Sp  q         q  r         r  t         t  Zu         _u  v         v  w         w  #y         (y  fz         kz  {         {  |         |  /~         4~  r         w                        ;         @  ~                    ƅ           	  G         L             ͊         Ҋ             S         X             ُ         ޏ           !  _           Ԕ         ٔ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     G                       N!  1          1  l=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -;-    ;-;-x    ;-#J-|    #J-q}-\    q}--    -q-h    q-b-    b--	    --h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               E!         V!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             >         ?  }         ~                        :                                              0            X  0            X            |                                                          8         9  w          x   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            A         C                        
           M         O                        
         
  Y         [                        "         $  e         g                        .         0  q         s                        :         <  }                                 F         H                          "         "  R#         T#  $         $  %         %  '         '  \(         `(  )         )  *         *  %,         ),  h-         l-  .         .  /         /  11         51  t2         x2  3         3  4         4  =6         A6  7         7  8         8  :         
:  I;         M;  <         <  =         =  ?         ?  U@         Y@  A         A  B         B  D         "D  aE         eE  F         F  G         G  *I          O  P          Q  Q          >R  %S          pS  WT          T  U         l  m         m  o         o  Qp         Sp  q         q  r         r  t         t  ]u         _u  v         v  w         w  &y         (y  iz         kz  {         {  |         |  2~         4~  u         w                        >         @             ą         ƅ           	  J         L             Њ         Ҋ             V         X             ܏         ޏ           !  b           ה         ٔ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     B                       :!  |1          1  I=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-;-x    ;-#J-|    #J-q}-\    q}--    --h    --b-    -b--	    --h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               C!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             >         ?  }         ~                        :                                              L  $          z  R                                                                                   8         9  w          x   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ?         C                                   K         O                        
         
  W         [                                  $  c         g                        ,         0  o         s                        8         <  {                                 D         H                          "         "  P#         T#  $         $  %         %  '         '  ^(         `(  )         )  *         *  ',         ),  j-         l-  .         .  /         /  31         51  v2         x2  3         3  4         4  ?6         A6  7         7  8         8  :         
:  K;         M;  <         <  =         =  ?         ?  W@         Y@  A         A  B         B   D         "D  cE         eE  F         F  G         G  ,I          [  \           ]  ]          2^  _          d_  H`          `  za         l  m         m  o         o  Op         Sp  q         q  r         r  t         t  [u         _u  v         v  w         w  $y         (y  gz         kz  {         {  |         |  0~         4~  s         w                        <         @                      ƅ           	  H         L             Ί         Ҋ             T         X             ڏ         ޏ           !  `           Ք         ٔ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     F                       N!  1          1  k=                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -A-    A-;-    ;-}J-    }J-}-d    }-1-    1--0    --    -Eb-H	    Eb-c-P
    c-+-
    +-9׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               F!         X!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 C	         D	  
         
                        *         +  i         j                                               F  "          u  Q                                              !              H!   "          o"  G#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q   w  operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       \            >         C                                   J         O                        
         
  V         [                                 $  b         g                        +         0  n         s                        7         <  z                                  C         H                          "         "  O#         T#  $         $  %         %  '         '  [(         `(  )         )  *         *  $,         ),  g-         l-  .         .  /         /  01         51  s2         x2  3         3  4         4  <6         A6  7         7  8         8  :         
:  H;         M;  <         <  =         =  ?         ?  T@         Y@  A         A  B         B  D         ng  h         h  i         i  2k         7k  ul         zl  m         m  n          o  >p         Cp  q         q  r         r  t         t  Ju         Ou  v          v  w          w  x          y  
z          Xz  L{          {  |         |  ~         ~  ]         b                        &         +  l         m                        2         3  t         u                        =         >           4  r         w                                b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     I                       X!  1          2  }=      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 c
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -5-    5-<-    <-J-    J-}-d    }-7-    7--0    --    -ib-H	    ib-c-P
    c-+-
    +-9׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               B!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         }                        9	         :	  x
         y
                        *         +  i         j                                               J  "          {  S                                  
            7   !          d!  <"          "  i#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q   w  operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       \            B         C                                   N         O                        
         
  Z         [                        #         $  f         g                        /         0  r         s                        ;         <  ~                                 G         H                          "         "  S#         T#  $         $  %         %  '         '  _(         `(  )         )  *         *  (,         ),  k-         l-  .         .  /         /  41         51  w2         x2  3         3  4         4  @6         A6  7         7  8         8  	:         
:  L;         M;  <         <  =         =  ?         ?  X@         Y@  A         A  B         B  !D         ng  h         h  i         i  6k         7k  yl         zl  m         m  n          o  Bp         Cp  q         q  r         r  t         t  Nu         Ou  v         v  w         w  y         y  Wz         Xz  {         {  |         |  ~         ~  a         b                        *          +            m  a                                  3  '         u                        9         >  |         4  v         w                                b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     M            !          l!  1          "2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 a
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -;-    ;-;-    ;-}J-    }J-}-d    }-7-    7--0    --    -?b-H	    ?b-c-P
    c-+-
    +-9׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               E!         V!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 C	         D	  
         
                        *         +  i         j                                               F  #          u  R                                              #              K!  #"          s"  K#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q   w  operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       \            ?         C                                   K         O                        
         
  W         [                                  $  c         g                        ,         0  o         s                        8         <  {                                 D         H                          "         "  P#         T#  $         $  %         %  '         '  \(         `(  )         )  *         *  %,         ),  h-         l-  .         .  /         /  11         51  t2         x2  3         3  4         4  =6         A6  7         7  8         8  :         
:  I;         M;  <         <  =         =  ?         ?  U@         Y@  A         A  B         B  D         ng  h         h  i         i  3k         7k  vl         zl  m         m  n          o  ?p         Cp  q         q  r         r  t         t  Ku         Ou  v         v  w         w  y         y  Tz         Xz  {         {  |          |  }          ~            b  W                        ݂         +  k         m                        1         3  s         u                        <         >           4  s         w                                b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     H                       T!  1          1  v=      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 d
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -o -`     o -ְ-     ְ-7ܰ-    7ܰ-4-x    4-@-\    @-i-<    i-7-    7-a-    a--    -]b- 	    ]b-c-(
    c-+-
    +-9׳-
                   =         b      J      7   U   7   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               C!         R!  P1                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 64 })  outputElements  5184    input0Elements  31104   input1Elements  24576   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     4                           #         b      J      7   M     operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       @         d                                  !  _	         $  b         c                                               Q  *            [                                              @            n  F                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     4                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  1728    input0Elements  1728    invocations     1                           3         b      J      7   Q   w  operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       \            A         C                        
           M         O                        
         
  Y         [                        "         $  e         g                        .         0  q         s                        :         <  }                                 F         H                          "         "  R#         T#  $         $  %         %  '         '  ^(         `(  )         )  *         *  ',         ),  j-         l-  .         .  /         /  31         51  v2         x2  3         3  4         4  ?6         A6  7         7  8         8  :         
:  K;         M;  <         <  =         =  ?         ?  W@         Y@  A         A  B         B   D         ng  h         h  i         i  5k         7k  xl         zl  m         m  n          o  Ap         Cp  q         q  r         r  
t         t  Mu         Ou  v         v  w         w  y         y  Vz         Xz  {         {  |         |  ~         ~  `         b                        )         +  i         m                        /         3  q          u  j                                  >  3         4  u         w                                b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     4                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                     L             !          h!  1          2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 b
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -S-    S-;-    ;-#J-    #J-q}-d    q}-%-    %--p    -b-    b-u-
    u--
    -հ-t    հ-?׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               J!         `!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             >         ?  }         ~                        :                                              -            T  ,          {  S            v                                                          .         /  m          n   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                      @  2            r                                  d#  $         $  %         %  "'         $'  b(         d(  )         )  *         *  ",         $,  b-         d-  .         .  /         /  "1         $1  b2         d2  3         3  4         4  "6         $6  b7         d7  8         8  9         9  ";         $;  b<         d<  =         =  >         >  "@         $@  bA         dA  B         B  C         C  "E         $E  bF         dF  G         G  H         H  "J         $J  cK         dK  L         L  M         M  #O         $O  cP         dP  Q         Q  R         R  #T         $T  cU         dU  V         V  W         W  #Y         $Y  cZ         dZ  [         [  \         \  #^         $^  c_         d_  `         `  a         a  #c         $c  cd         dd  e         e  f         f  #h         $h  ci         di  j         j  k         k  #m         $m  cn         dn  o         ds  t         t  u         u  "w         $w  bx         dx  y         y  z         z  "|         $|  b}         d}  ~         ~             "         $  b         d                        "         $  b         d                        "         $  b         d                        "         $  b         d                        "         $  b         d                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               O!         j!  2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
                   
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     &
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -G-    G-;-x    ;-#J-|    #J-q}-\    q}-2-    2--h    -b-    b-	n-	    	n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               F!         X!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             >         ?  }         ~                        :                                              I  !          v  N            {                                                                      .         /  m          n   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ?         @                                    ?         @                        	          
  ?         @                                    ?         @                                    ?         @                                    ?         @                                    ?         @               !          d#  G$          $  u%          %  &          &  '          (  (         F  H         H  GI         I  J         J  L         L  QM         SM  N         N  O         O  Q         Q  QR         SR  S         S  T         T  V         V  QW         SW  X         X  Y         Y  [         [  Q\         S\  ]         ]  ^         ^  `         `  Qa         Sa  b         b  c         c  e         e  Qf         Sf  g         g  h         h  j         j  Qk         Sk  l         l  m         m  o         o  p         p  #r         $r  cs           א         ؐ             W         X             ו         ؕ             W         X             ך         ؚ             W         X             ן         ؟             W         X             פ         ؤ             W         X             ש         ة             W         X             ׮         خ             W         X                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               K!         b!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     )
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -M-    M-;-    ;-/J-    /J-}}-d    }}-%-    %--p    -b-    b-u-
    u--
    -ϰ-t    ϰ-?׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               H!         \!  1         
2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             >         ?  }         ~                        :                                              0            X  0            X            |                                                          8         9  w          x   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         @  2              r!         d#  $         $  %         %  #'         $'  c(         d(  )         )  *         *  #,         $,  c-         d-  .         .  /         /  #1         $1  c2         d2  3         3  4         4  #6         $6  c7         d7  8         8  9         9  #;         $;  c<         d<  =         =  >         >  #@         $@  cA         dA  B         B  C         C  #E         $E  cF         dF  G         G  H         H  #J         $J  bK         dK  L         L  M         M  "O         $O  bP         dP  Q         Q  R         R  "T         $T  bU         dU  V         V  W         W  "Y         $Y  bZ         dZ  [         [  \         \  "^         $^  b_         d_  `         `  a         a  "c         $c  bd         dd  e         e  f         f  "h         $h  bi         di  j         j  k         k  "m         $m  bn         dn  o          o  p          p  q          $r  s         ds  t         t  u         u  #w         $w  cx         dx  y         y  z         z  #|         $|  c}         d}  ~         ~             #         $  c         d                        #         $  c         d                        #         $  c         d                        #         $  c         d                        #         $  c         d                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               O!         j!  2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
                   
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     %
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -G-    G-;-x    ;-/J-|    /J-}}-\    }}-2-    2--h    -b-    b-	n-	    	n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               F!         X!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             >         ?  }         ~                        :                                              L  $          z  R                                                                                   8         9  w          x   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            >         @  ~                                  >         @  ~                      	          
  >         @  ~                                  >         @  ~                                  >         @  ~                                  >         @  ~                                  >         @  ~             !          J)  -*          x*  [+          +  ,          ,  -          .  .         F  H         H  FI         I  J         J  L         L  RM         SM  N         N  O         O  Q         Q  RR         SR  S         S  T         T  V         V  RW         SW  X         X  Y         Y  [         [  R\         S\  ]         ]  ^         ^  `         `  Ra         Sa  b         b  c         c  e         e  Rf         Sf  g         g  h         h  j         j  Rk         Sk  l         l  m         m  o         o  p         p  "r         $r  bs           ֐         ؐ             V         X             ֕         ؕ             V         X             ֚         ؚ             V         X             ֟         ؟             V         X             ֤         ؤ             V         X             ֩         ة             V         X             ֮         خ             V         X                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               M!         f!   2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     *
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -G-    G-6-x    6-IE-|    IE-x-\    x--    --h    -b-    b-n-	    n--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               F!         X!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         #            F            i  A            d                                              B	  $
          s
  U                      !         "  `         a                                   Q         R                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            >         C                                   J         O                        
         
  V         [                                 $  b         g                        +         0  n         s                        7         <  z                                  C         H                          "         "  O#         T#  $         $  %         %  '         '  _(         `(  )         )  *         *  (,         ),  k-         l-  .         .  /         /  41         51  w2         x2  3         3  4         4  @6         A6  7         7  8         8  	:         
:  L;         M;  <         <  =         =  ?         ?  X@         Y@  A         A  B         B  !D         "D  dE         eE  F          F  G          G  H          H  I          J  J          @K  L         i  j         j  ;l         <l  ~m         m  n         n  p         p  Gq         Hq  r         r  s         s  u         u  Sv         Tv  w         w  x         x  z         z  _{         `{  |         |  }         }  (         )  k         l                        4         5  w         x                        @         A             Ƌ         ǋ  	         
  L         M             Ґ         Ӑ             X         Y                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               F!         X!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     +
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -;-    ;-6-x    6-E-|    E-gx-\    gx--    --h    -b-    b-n-	    n--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               B!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     :            W  /          t  L            w                                  	  
          M
  3         |                        8         9  w         x                                    ^         _                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            B         C                                   N         O                        
         
  Z         [                        #         $  f         g                        /         0  r         s                        ;         <  ~                                 G         H                          "         "  S#         T#  $         $  %         %  '         '  [(         `(  )         )  *         *  $,         ),  g-         l-  .         .  /         /  01         51  s2         x2  3         3  4         4  <6         A6  7         7  8         8  :         
:  H;         M;  <         <  =         =  ?         ?  T@         Y@  A         A  B         B  D         "D  `E         eE  F          $R  R          GS  T          jT  BU          U  eV          V  W         i  j         j  7l         <l  zm         m  n         n   p         p  Cq         Hq  r         r  s         s  u         u  Ov         Tv  w         w  x         x  z         z  [{         `{  |         |  }         }  $         )  g         l                        0         5  s         x                        <         A                      ǋ           
  H         M             ΐ         Ӑ             T         Y                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               B!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     )
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -A-    A-6-x    6-OE-|    OE-x-\    x--    --h    -b-    b-n-	    n--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               D!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         $            H             l  D            h                                              G	  *
          x
  [                      &         '  e         f                                   V         W                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ?         C                                   K         O                        
         
  W         [                                  $  c         g                        ,         0  o         s                        8         <  {                                 D         H                          "         "  P#         T#  $         $  %         %  '         '  ^(         `(  )         )  *         *  ',         ),  j-         l-  .         .  /         /  31         51  v2         x2  3         3  4         4  ?6         A6  7         7  8         8  :         
:  K;         M;  <         <  =         =  ?         ?  W@         Y@  A         A  B         B   D         "D  cE         eE  F          fL  >M          M  dN          N  O          O  P          P  Q         i  j         j  :l         <l  }m         m  n         n  p         p  Fq         Hq  r         r  s         s  u         u  Rv         Tv  w         w  x         x  z         z  ^{         `{  |         |  }         }  '         )  j         l                        3         5  v         x                        ?         A             ŋ         ǋ           
  K         M             ѐ         Ӑ             W         Y                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               D!         T!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     *
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -;-    ;-6-x    6-E-|    E-mx-\    mx--    --h    -b-    b-n-	    n--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               B!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                     <            Z  2          x  P            }                                  #	  

          R
  9                                 =         >  |         }                        $         %  c         d                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            A         C                        
           M         O                        
         
  Y         [                        "         $  e         g                        .         0  q         s                        :         <  }                                 F         H                          "         "  R#         T#  $         $  %         %  '         '  \(         `(  )         )  *         *  %,         ),  h-         l-  .         .  /         /  11         51  t2         x2  3         3  4         4  =6         A6  7         7  8         8  :         
:  I;         M;  <         <  =         =  ?         ?  U@         Y@  A         A  B         B  D         "D  aE         eE  F          W  X          X  Y          Z  Z          <[  \          _\  7]         i  j         j  8l         <l  {m         m  n         n  p         p  Dq         Hq  r         r  s         s  u         u  Pv         Tv  w         w  x         x  z         z  \{         `{  |         |  }         }  %         )  h         l                        1         5  t         x                        =         A             Ë         ǋ           
  I         M             ϐ         Ӑ             U         Y                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               B!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     *
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -M-    M-;-x    ;-J-|    J-}-\    }-i6-    i6--h    -b-    b-kx-	    kx--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               F!         X!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 C	         D	  
         
                        *         +  i         j                                               F  "          u  Q                                              !              H!   "          o"  G#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            >         C                                   J         O                        
         
  V         [                                 $  b         g                        +         0  n         s                        7         <  z                                  C         H                          "         "  O#         T#  $         $  %         F  G         G  )I         .I  lJ         qJ  K         K  L         L  5N         :N  xO         }O  P         P  Q         R  AS         FS  T         T  U         U  
W         W  MX         RX  Y         Y  Z         Z  \         \  Y]         ^]  ^         ^  _         _  "a         'a  eb         jb  c         c  d         d  .f         3f  qg         vg  h         h  i         i  :k         ?k  }l         l  m         m  o         o  Jp         Kp  q         q  r         r  t         t  Vu         Wu  v         v  w         w  y          y  bz         cz  {         {  |         |  +~         ,~  n         o                        7         8  z         {                         C         D             Ɋ         ʊ             O          v  a                        ݞ          0            n  Y         ұ             W         X             ݶ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               I!         ^!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 c
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -G-    G-<-x    <-J-|    J-}-\    }-c6-    c6--h    -b-    b-qx-	    qx--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               F!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         }                        9	         :	  x
         y
                        *         +  i         j                                               J  "          {  S                                  
            7   !          d!  <"          "  i#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            B         C                                   N         O                        
         
  Z         [                        #         $  f         g                        /         0  r         s                        ;         <  ~                                 G         H                          "         "  S#         T#  $         $  %         F  G         G  -I         .I  pJ         qJ  K         K  L         L  9N         :N  |O         }O  P         P  R         R  ES         FS  T         T  U         U  W         W  QX         RX  Y         Y  Z         Z  \         \  ]]         ^]  ^         ^  _         _  &a         'a  ib         jb  c         c  d         d  2f         3f  ug         vg  h         h  i         i  >k         ?k  l         l  m         m  o         o  Fp         Kp  q         q  r         r  t         t  Ru         Wu  v         v  w         w  y          y  ^z         cz  {         {  |         |  '~         ,~  j         o                        3         8  v         {                        ?         D             Ŋ         ʊ             K            Ǩ                       ^  C                      ڬ           ұ             S         X             ٶ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               E!         V!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 e
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -G-    G-;-x    ;-J-|    J-}-\    }-i6-    i6--h    -b-    b-kx-	    kx--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               D!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 C	         D	  
         
                        *         +  i         j                                               F  #          u  R                                              #              K!  #"          s"  K#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ?         C                                   K         O                        
         
  W         [                                  $  c         g                        ,         0  o         s                        8         <  {                                 D         H                          "         "  P#         T#  $         $  %         F  G         G  *I         .I  mJ         qJ  K         K  L         L  6N         :N  yO         }O  P         P  Q         R  BS         FS  T         T  U         U  W         W  NX         RX  Y         Y  Z         Z  \         \  Z]         ^]  ^         ^  _         _  #a         'a  fb         jb  c         c  d         d  /f         3f  rg         vg  h         h  i         i  ;k         ?k  ~l         l  m         m  o         o  Ip         Kp  q         q  r         r  t         t  Uu         Wu  v         v  w         w  y          y  az         cz  {         {  |         |  *~         ,~  m         o                        6         8  y         {                        B         D             Ȋ         ʊ             N                        գ          (            f  Q                     ұ             V         X             ܶ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               I!         ^!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 c
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -A-    A-<-x    <-J-|    J-}-\    }-8-    8--h    -b-    b-qx-	    qx--h
    -E׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               B!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         }                        9	         :	  x
         y
                        *         +  i         j                                               J  #          {  T                                              9   !          g!  ?"          "  m#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            A         C                        
           M         O                        
         
  Y         [                        "         $  e         g                        .         0  q         s                        :         <  }                                 F         H                          "         "  R#         T#  $         $  %         F  G         G  ,I         .I  oJ         qJ  K         K  L         L  8N         :N  {O         }O  P         P  R         R  DS         FS  T         T  U         U  W         W  PX         RX  Y         Y  Z         Z  \         \  \]         ^]  ^         ^  _         _  %a         'a  hb         jb  c         c  d         d  1f         3f  tg         vg  h         h  i         i  =k         ?k  l         l  m         m  o         o  Gp         Kp  q         q  r         r  t         t  Su         Wu  v         v  w         w  y          y  _z         cz  {         {  |         |  (~         ,~  k         o                        4         8  w         {                        @         D             Ɗ         ʊ  	           L                      V  ;            y         ұ             T         X             ڶ          ޶  ÷                               >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               G!         Z!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 e
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -S-    S-6-x    6-aE-|    aE-x-\    x--    --h    -b-    b-qx-	    qx--h
    -K׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               H!         X!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         &            L  $          r  J            p                                     	          Q	  8
          
  i                      0         1  o         p                        !         "  `         a                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         `  J                                              @  *         %  2'         7'  u(         z(  )         )  *          +  >,         C,  -         -  .         .  0         0  J1         O1  2         2  3         3  5         5  V6         [6  7         7  8         8  :         $:  b;         g;  <         <  =         =  +?         0?  n@         s@  A         A  B         B  7D         <D  zE         E  F         F   H         H  CI         HI  J         J  K         K  M         M  SN         TN  O         O  P         P  R         R  _S         `S  T         T  U         U  (W         )W  kX         lX  Y         Y  Z         Z  4\         5\  w]         x]  ^         ^  _         _  @a         Aa  b         b  c         c  	e         
e  Lf         Mf  g         g  h         h  j         j  Xk         Yk  l         l  m         m  !o         "o  dp         ep  q         q  r         r  -t         .t  pu         qu  v         v  w         w  9y         :y  |z         }z  {         {  }         }  E~         F~             ˀ         ̀             Q         R             ׅ         ؅             ]         ^                        &         '  i         j                        2         3  u         v                        >         ?                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               D!         T!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 c
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -M-    M-6-x    6-7E-|    7E-x-\    x--    --h    -b-    b-qx-	    qx--h
    -K׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               H!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         "            D            f  >            `                                              7	  
          f
  G                                 Q         R                                   8         9  w         x                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                        C  8            {                                 %  6'         7'  y(         z(  )         )  *          +  B,         C,  -         -  .         .  0         0  N1         O1  2         2  3         3  5         5  Z6         [6  7         7  8         8  #:         $:  f;         g;  <         <  =         =  /?         0?  r@         s@  A         A  B         B  ;D         <D  ~E         E  F         F  H         H  GI         HI  J         J  K         K  M         M  ON         TN  O         O  P         P  R         R  [S         `S  T         T  U         U  $W         )W  gX         lX  Y         Y  Z         Z  0\         5\  s]         x]  ^         ^  _         _  <a         Aa  b         b  c         c  e         
e  Hf         Mf  g         g  h         h  j         j  Tk         Yk  l         l  m         m  o         "o  `p         ep  q         q  r         r  )t         .t  lu         qu  v         v  w         w  5y         :y  xz         }z  {         {  |         }  A~         F~             ǀ         ̀  
           M         R             Ӆ         ؅             Y         ^             ߊ           "         '  e         j                        .         3  q         v                        :         ?  }                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               @!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 e
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -M-    M-6-x    6-[E-|    [E-x-\    x--    --h    -b-    b-qx-	    qx--h
    -K׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               F!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         %            J  "          o  G            l                                    	          L	  2
          }
  c                      +         ,  j         k                                   [         \                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         x  b                          !           "  
#          X#  B$         %  3'         7'  v(         z(  )         )  *          +  ?,         C,  -         -  .         .  0         0  K1         O1  2         2  3         3  5         5  W6         [6  7         7  8         8   :         $:  c;         g;  <         <  =         =  ,?         0?  o@         s@  A         A  B         B  8D         <D  {E         E  F         F  H         H  DI         HI  J         J  K         K  M         M  RN         TN  O         O  P         P  R         R  ^S         `S  T         T  U         U  'W         )W  jX         lX  Y         Y  Z         Z  3\         5\  v]         x]  ^         ^  _         _  ?a         Aa  b         b  c         c  e         
e  Kf         Mf  g         g  h         h  j         j  Wk         Yk  l         l  m         m   o         "o  cp         ep  q         q  r         r  ,t         .t  ou         qu  v         v  w         w  8y         :y  {z         }z  {         {  }         }  D~         F~             ʀ         ̀             P         R             օ         ؅             \         ^                        %         '  h         j                        1         3  t         v                        =         ?                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               B!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 c
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -G-    G-6-x    6-=E-|    =E-x-\    x--    --h    -b-    b-qx-	    qx--h
    -K׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               D!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         #            F            i  A            d                                              <	  
          k
  M                                 V         W                                   =         >  |         }                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         O  D                        	          
            [  P         %  5'         7'  x(         z(  )         )  *          +  A,         C,  -         -  .         .  
0         0  M1         O1  2         2  3         3  5         5  Y6         [6  7         7  8         8  ":         $:  e;         g;  <         <  =         =  .?         0?  q@         s@  A         A  B         B  :D         <D  }E         E  F         F  H         H  FI         HI  J         J  K         K  M         M  PN         TN  O         O  P         P  R         R  \S         `S  T         T  U         U  %W         )W  hX         lX  Y         Y  Z         Z  1\         5\  t]         x]  ^         ^  _         _  =a         Aa  b         b  c         c  e         
e  If         Mf  g         g  h         h  j         j  Uk         Yk  l         l  m         m  o         "o  ap         ep  q         q  r         r  *t         .t  mu         qu  v         v  w         w  6y         :y  yz         }z  {         {  |         }  B~         F~             Ȁ         ̀             N         R             ԅ         ؅             Z         ^                        #         '  f         j                        /         3  r         v                        ;         ?  ~                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               @!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 e
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     ---X     ---     -Y-    Y-I<-x    I<-J-|    J-1~-\    1~-8-    8--h    -b-    b-߂-	    ߂--h
    -Q׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               O!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 C	         D	  
         
                        *         +  i         j                                               T  ,            b                                  (              Z   2!          !  d"          "  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            >         C                                   J         O                        
         
  V         [                                 $  b         g                        +         0  n         s                        7         <  z                                  C         H                          "         "  O#         T#  $         $  %         %  '         '  [(         `(  )         )  *         *  $,         ),  g-         l-  .         .  /         /  01         51  s2         x2  3         3  4         4  <6         A6  7         7  8         8  :         
:  H;         M;  <         <  =         =  ?         ?  T@         Y@  A         A  B         B  D         "D  `E         eE  F         F  G         G  )I         .I  lJ         qJ  K          K  L          L  M          $N  N          WO  /P          P  bQ         :  |         }                        E         F             ˚         ̚             Q         R             ן         ؟             ]         ^                        &         '  i         j                        2         3  u         v                        >         ?             Ĳ         Ų             J         K             з         ѷ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               >!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -A-    A-C<-x    C<-J-|    J-+~-\    +~-8-    8--h    -b-    b-߂-	    ߂--h
    -Q׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               C!         D!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         }                        9	         :	  x
         y
                        *         +  i         j                                               L  $          ~  V                                              N   &!          !  \"          "  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            B         C                                   N         O                        
         
  Z         [                        #         $  f         g                        /         0  r         s                        ;         <  ~                                 G         H                          "         "  S#         T#  $         $  %         %  '         '  _(         `(  )         )  *         *  (,         ),  k-         l-  .         .  /         /  41         51  w2         x2  3         3  4         4  @6         A6  7         7  8         8  	:         
:  L;         M;  <         <  =         =  ?         ?  X@         Y@  A         A  B         B  !D         "D  dE         eE  F         F  G         G  -I         .I  pJ         qJ  K          W  X          X  Y          (Z   [          ^[  6\          \  l]         :  x         }                        A         F             ǚ         ̚  
           M         R             ӟ         ؟             Y         ^             ߤ           "         '  e         j                        .         3  q         v                        :         ?  }                    Ų             F         K             ̷         ѷ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               B!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     ---X     ---     -Y-    Y-I<-x    I<-J-|    J-1~-\    1~-8-    8--h    -b-    b-߂-	    ߂--h
    -Q׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               O!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 C	         D	  
         
                        *         +  i         j                                               R  *            _                                  $            W   /!          !  b"          "  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ?         C                                   K         O                        
         
  W         [                                  $  c         g                        ,         0  o         s                        8         <  {                                 D         H                          "         "  P#         T#  $         $  %         %  '         '  \(         `(  )         )  *         *  %,         ),  h-         l-  .         .  /         /  11         51  t2         x2  3         3  4         4  =6         A6  7         7  8         8  :         
:  I;         M;  <         <  =         =  ?         ?  U@         Y@  A         A  B         B  D         "D  aE         eE  F         F  G         G  *I         .I  mJ         qJ  K          Q  R          R  S          #T  T          VU  .V          V  aW         :  {         }                        D         F             ʚ         ̚             P         R             ֟         ؟             \         ^                        %         '  h         j                        1         3  t         v                        =         ?             ò         Ų             I         K             Ϸ         ѷ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               >!         H!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -G-    G-C<-x    C<-J-|    J-+~-\    +~-8-    8--h    -b-    b-߂-	    ߂--h
    -Q׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               G!         H!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         }                        9	         :	  x
         y
                        *         +  i         j                                               N  &            Y                                              Q   )!          !  ^"          "  #                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            A         C                        
           M         O                        
         
  Y         [                        "         $  e         g                        .         0  q         s                        :         <  }                                 F         H                          "         "  R#         T#  $         $  %         %  '         '  ^(         `(  )         )  *         *  ',         ),  j-         l-  .         .  /         /  31         51  v2         x2  3         3  4         4  ?6         A6  7         7  8         8  :         
:  K;         M;  <         <  =         =  ?         ?  W@         Y@  A         A  B         B   D         "D  cE         eE  F         F  G         G  ,I         .I  oJ         qJ  K          ]  ^           _  _          6`  a          la  Db          b  zc         :  y         }                        B         F             Ț         ̚             N         R             ԟ         ؟             Z         ^                        #         '  f         j                        /         3  r         v                        ;         ?  ~                    Ų             G         K             ͷ         ѷ                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               @!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     ---X     ---     -_-    _-;-x    ;-5J-|    5J-}-\    }-k-    k--h    -ub-    ub-ق-	    ق--h
    -Q׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               U!         V!   2         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             >         ?  }         ~                        :                                                          4            S  +          ~  V                                                          .         /  m          n   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            B         C                                   N         O                        
         
  Z         [                        #         $  f         g                        /         0  r         s                        ;         <  ~                                 G         H                          "         "  S#         T#  $         $  %         %  '         '  [(         `(  )         )  *         *  $,         ),  g-         l-  .         .  /         /  01         51  s2         x2  3         3  4         4  <6         A6  7         7  8         8  :         
:  H;         M;  <         <  =         =  ?         ?  T@         Y@  A         A  B         B  D         "D  `E         eE  F         F  G         G  )I          T  oU          U  V          V  W          X  X          &Y  Y         zk  l         l  m          n  Bo         Co  p         p  q         q  s         s  Nt         Ot  u         u  v         v  x         x  Zy         [y  z         z  {         {  #}         $}  f~         g~                        /         0  r         s                        ;         <  ~                                 G         H             ͎         Ύ             S         T             ٓ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               5!         6!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     ---X     ---     -Y-    Y-;-x    ;-5J-|    5J-}-\    }-e-    e--h    -b-    b-ӂ-	    ӂ--h
    -Q׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               Q!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             >         ?  }         ~                        :                                              1  	          V  .          {  S            x                                                          .         /  m          n   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            >         C                                   J         O                        
         
  V         [                                 $  b         g                        +         0  n         s                        7         <  z                                  C         H                          "         "  O#         T#  $         $  %         %  '         '  _(         `(  )         )  *         *  (,         ),  k-         l-  .         .  /         /  41         51  w2         x2  3         3  4         4  @6         A6  7         7  8         8  	:         
:  L;         M;  <         <  =         =  ?         ?  X@         Y@  A         A  B         B  !D         "D  dE         eE  F         F  G         G  -I          _  `          $a  a          Jb  "c          pc  Hd          d  ne         zk  l         l  m          n  >o         Co  p         p  q         q  s         s  Jt         Ot  u         u  v         v  x         x  Vy         [y  z         z  {         {  }         $}  b~         g~                        +         0  n         s                        7         <  z                                  C         H             Ɏ         Ύ             O         T             Փ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               9!         >!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     ---X     ---     -Y-    Y-;-x    ;-;J-|    ;J-}-\    }-k-    k--h    -{b-    {b-ق-	    ق--h
    -Q׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               Q!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             >         ?  }         ~                        :                                                          8            X  0            \                                                          8         9  w          x   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            A         C                        
           M         O                        
         
  Y         [                        "         $  e         g                        .         0  q         s                        :         <  }                                 F         H                          "         "  R#         T#  $         $  %         %  '         '  \(         `(  )         )  *         *  %,         ),  h-         l-  .         .  /         /  11         51  t2         x2  3         3  4         4  =6         A6  7         7  8         8  :         
:  I;         M;  <         <  =         =  ?         ?  U@         Y@  A         A  B         B  D         "D  aE         eE  F         F  G         G  *I          JZ  #[          n[  G\          \  k]          ]  ^          ^  _         zk  l         l  m          n  Ao         Co  p         p  q         q  
s         s  Mt         Ot  u         u  v         v  x         x  Yy         [y  z         z  {         {  "}         $}  e~         g~                        .         0  q         s                        :         <  }                                 F         H             ̎         Ύ             R         T             ؓ                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               7!         :!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     ---X     ---     -Y-    Y-;-x    ;-;J-|    ;J-}-\    }-k-    k--h    -b-    b-ӂ-	    ӂ--h
    -Q׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               Q!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             >         ?  }         ~                        :                                              4            Z  2            X            ~                                                          8         9  w          x   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ?         C                                   K         O                        
         
  W         [                                  $  c         g                        ,         0  o         s                        8         <  {                                 D         H                          "         "  P#         T#  $         $  %         %  '         '  ^(         `(  )         )  *         *  ',         ),  j-         l-  .         .  /         /  31         51  v2         x2  3         3  4         4  ?6         A6  7         7  8         8  :         
:  K;         M;  <         <  =         =  ?         ?  W@         Y@  A         A  B         B   D         "D  cE         eE  F         F  G         G  ,I          e  f          f  g          h  h          .i  j          Tj  ,k         zk  l         l  m          n  ?o         Co  p         p  q         q  s         s  Kt         Ot  u         u  v         v  x         x  Wy         [y  z         z  {         {   }         $}  c~         g~                        ,         0  o         s                        8         <  {                                 D         H             ʎ         Ύ             P         T             ֓                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               9!         >!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -M-    M-;-x    ;-kJ-|    kJ-}-\    }-/-    /--P    -b-    b-c-	    c-7-D
    7-K׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               L!         P!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 C	         D	  
         
                        *         +  i         j                                               F  *          u  Y                                                            0!  "          O"  '#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            >         C                                   J         O                        
         
  V         [                                 $  b         g                        +         0  n         s                        7         <  z                                  C         H                          "         2  3         3  	5         5  L6         Q6  7         7  8         8  :         :  X;         ];  <         <  =         =  !?         &?  d@         i@  A         A  B         B  -D         2D  pE         uE  F         F  G         G  9I         >I  |J         J  K         K  M         M  EN         JN  O         O  P         P  R         R  QS         VS  T         T  U         U  W          `  a          a  b          'c  d          [d  De          e  xf         l  n         n  Uo         Zo  p         p  q         q   s         !s  bt         ct  u         u  v         v  *x         +x  my         ny  z         z  {         {  6}         7}  y~         z~                         B         C             Ȅ         Ʉ             N         O             ԉ         Չ             Z         [                        #         v                        >                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               D!         T!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 a
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -S-    S-;-    ;-J-    J-}-d    }-/-    /-e7-0    e7--    -b-H	    b-c-P
    c-7-
    7-K׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               L!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         }                        9	         :	  x
         y
                        *         +  i         j                                               J  *          {  [                                              '              L!  $"          q"  I#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q   w  operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       \            B         C                                   N         O                        
         
  Z         [                        #         $  f         g                        /         0  r         s                        ;         <  ~                                 G         H                          "         2  3         3  5         5  P6         Q6  7         7  8         8  :         :  \;         ];  <         <  =         =  %?         &?  h@         i@  A         A  B         B  1D         2D  tE         uE  F         F  G         G  =I         >I  J         J  K         K  M         M  IN         JN  O         O  P         P  R         R  US         VS  T         T  U         U  W          l  m         n  Yo         Zo  p          p  q          q  r          !s  t          ct  Wu         u  v         v  &x         +x  iy         ny  z         z  {         {  2}         7}  u~         z~                         >         C             Ą         Ʉ             J         O             Љ         Չ             V         [             ܎                    v                        :                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               H!         \!  1         
2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 _
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -G-    G-;-x    ;-qJ-|    qJ-}-\    }-/-    /--P    -b-    b-c-	    c-7-D
    7-K׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               I!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 C	         D	  
         
                        *         +  i         j                                               F  +          u  Z                                                            3!  "          S"  +#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       p            ?         C                                   K         O                        
         
  W         [                                  $  c         g                        ,         0  o         s                        8         <  {                                 D         H                          "         2  3         3  
5         5  M6         Q6  7         7  8         8  :         :  Y;         ];  <         <  =         =  "?         &?  e@         i@  A         A  B         B  .D         2D  qE         uE  F         F  G         G  :I         >I  }J         J  K         K  M         M  FN         JN  O         O  P         P  R         R  RS         VS  T         T  U         U  W          f  g          g  h          +i  j         l  n         n  Vo         Zo  p         p  q         q  s         !s  at         ct  u         u  v         v  )x         +x  ly         ny  z         z  {         {  5}         7}  x~         z~                         A         C             Ǆ         Ʉ  
           M         O             Ӊ         Չ             Y         [             ߎ           "          T  =            q         v                        =                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               D!         T!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              	                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              
                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 a
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -M-    M-;-    ;-J-    J-}-d    }-/-    /-e7-0    e7--    -b-H	    b-c-P
    c-7-
    7-K׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               K!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         }                        9	         :	  x
         y
                        *         +  i         j                                               J  +          {  \                                              )   !          O!  '"          u"  M#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q   w  operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       \            A         C                        
           M         O                        
         
  Y         [                        "         $  e         g                        .         0  q         s                        :         <  }                                 F         H                          "         2  3         3  5         5  O6         Q6  7         7  8         8  :         :  [;         ];  <         <  =         =  $?         &?  g@         i@  A         A  B         B  0D         2D  sE         uE  F         F  G         G  <I         >I  J         J  K         K  M         M  HN         JN  O         O  P         P  R         R  TS         VS  T         T  U         U  W         l  n         n  Xo         Zo  p         p  q         q  s         !s  _t         ct  u          u  v          v  w          +x   y          ny  cz          z  {         {  3}         7}  v~         z~                         ?         C             ń         Ʉ             K         O             щ         Չ             W         [             ݎ                     v                        ;                              b            W   u      operation.2     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   6       bytes   864     pattern Contiguous      invocations     5                           >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               F!         X!  1         2  =      	                        b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  37584   input1Elements  29696   invocations     1              
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 _
                               b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -A-    A-;-    ;-)J-    )J-w}-d    w}--    --p    -b-    b-{-
    {--
    -ᰳ-t    ᰳ-E׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               G!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             >         ?  }         ~                        :                                                          4            S  +          ~  V                                                          .         /  m          n   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                        <  *          x  f                                 %  F&         K&  '         '  (         (  *         *  R+         W+  ,         ,  -         -  /          /  ^0         c0  1         1  2         2  '4         ,4  j5         o5  6         6  7         7  39         89  v:         {:  ;         ;  <         =  ?>         D>  ?         ?  @         @  B         B  KC         PC  D         D  E         E  G         G  WH         \H  I         I  J         J   L         %L  cM         hM  N         N  O         O  ,Q         1Q  oR         tR  S         S  T         T  8V         =V  {W         W  X         X  Z         Z  D[         I[  \         \  ]         ]  _         _  P`         U`  a         a  b         b  d         d  \e         ae  f         f  g         g  %i         *i  hj         mj  k         k  l         l  5n         6n  xo         yo  p         p  q         q  As         Bs  t         t  u         u  
w         w  Mx         Nx  y         y  z         z  |         |  Y}         Z}  ~         ~             "         #  e         f                        .         /  q         r                        :         ;  }         ~                        F         G             ̒         ͒             N         S                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               O!         b!   2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
                   
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     .
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -A-    A-;-    ;-)J-    )J-w}-d    w}--    --p    -b-    b-{-
    {--
    -۰-t    ۰-E׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               G!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             >         ?  }         ~                        :                                              1  	          V  .          {  S            x                                                          .         /  m          n   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         X  I                                              H  9         %  J&         K&  '         '  (         (  *         *  V+         W+  ,         ,  -         -  /          /  b0         c0  1         1  2         2  +4         ,4  n5         o5  6         6  7         7  79         89  z:         {:  ;         ;   =         =  C>         D>  ?         ?  @         @  B         B  OC         PC  D         D  E         E  G         G  [H         \H  I         I  J         J  $L         %L  gM         hM  N         N  O         O  0Q         1Q  sR         tR  S         S  T         T  <V         =V  W         W  X         X  Z         Z  H[         I[  \         \  ]         ]  _         _  T`         U`  a         a  b         b  d         d  `e         ae  f         f  g         g  )i         *i  lj         mj  k         k  l         l  1n         6n  to         yo  p         p  q         q  =s         Bs  t         t  u         u  w         w  Ix         Nx  y         y  z         z  |         |  U}         Z}  ~         ~                      #  a         f                        *         /  m         r                        6         ;  y         ~                        B         G             Ȓ         ͒             R         S                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               O!         f!  2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
                   
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     &
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -;-    ;-;-    ;-5J-    5J-}-d    }--    --p    -b-    b-{-
    {--
    -ᰳ-t    ᰳ-E׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               E!         J!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             >         ?  }         ~                        :                                                          8            X  0            \                                                          8         9  w          x   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         ,            h  V            	          	  
            
         %  G&         K&  '         '  (         (  *         *  S+         W+  ,         ,  -         -  /          /  _0         c0  1         1  2         2  (4         ,4  k5         o5  6         6  7         7  49         89  w:         {:  ;         ;  <         =  @>         D>  ?         ?  @         @  	B         B  LC         PC  D         D  E         E  G         G  XH         \H  I         I  J         J  !L         %L  dM         hM  N         N  O         O  -Q         1Q  pR         tR  S         S  T         T  9V         =V  |W         W  X         X  Z         Z  E[         I[  \         \  ]         ]  _         _  Q`         U`  a         a  b         b  d         d  ]e         ae  f         f  g         g  &i         *i  ij         mj  k         k  l         l  4n         6n  wo         yo  p         p  q         q  @s         Bs  t         t  u         u  	w         w  Lx         Nx  y         y  z         z  |         |  X}         Z}  ~         ~             !         #  d         f                        -         /  p         r                        9         ;  |         ~                        E         G             ˒         ͒             O         S                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               M!         b!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
                   
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     -
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    4     -,-`     ,--     -A-    A-;-    ;-5J-    5J-}-d    }--    --p    -b-    b-{-
    {--
    -۰-t    ۰-E׳-                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               G!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             >         ?  }         ~                        :                                              4            Z  2            X            ~                                                          8         9  w          x   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |           u                                  8  )          t  e         %  I&         K&  '         '  (         (  *         *  U+         W+  ,         ,  -         -  /          /  a0         c0  1         1  2         2  *4         ,4  m5         o5  6         6  7         7  69         89  y:         {:  ;         ;  <         =  B>         D>  ?         ?  @         @  B         B  NC         PC  D         D  E         E  G         G  ZH         \H  I         I  J         J  #L         %L  fM         hM  N         N  O         O  /Q         1Q  rR         tR  S         S  T         T  ;V         =V  ~W         W  X         X  Z         Z  G[         I[  \         \  ]         ]  _         _  S`         U`  a         a  b         b  d         d  _e         ae  f         f  g         g  (i         *i  kj         mj  k         k  l         l  2n         6n  uo         yo  p         p  q         q  >s         Bs  t         t  u         u  w         w  Jx         Nx  y         y  z         z  |         |  V}         Z}  ~         ~                      #  b         f                        +         /  n         r                        7         ;  z         ~                         C         G             ɒ         ͒             Q         S                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               O!         f!  2         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
                   
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     %
                             b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    8     -,-h     ,--     -;-    ;-6-    6-sE-    sE-x-l    x-{9-    {9--x    -b-	    b-{-
    {--
    --|    -%ֳ-    %ֳ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               E!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         +            V  .            Y                                              9  	          j	  D
          
  u           
           I         J                                   :         ;  y         z                        6                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |        D%  &          ?2  3          k3  I4          4  u5          5  6          6  7         I  J         J  4L         5L  wM         xM  N         N  O         O  @Q         AQ  R         R  S         S  	U         
U  LV         MV  W         W  X         X  Z         Z  X[         Y[  \         \  ]         ]  !_         "_  d`         e`  a         a  b         b  -d         .d  pe         qe  f         f  g         g  9i         :i  |j         }j  k         k  m         m  An         Fn  o         o  p         p  r         r  Qs         Rs  t         t  u         u  w         w  ]x         ^x  y         y  z         z  &|         '|  i}         j}  ~         ~             2         3  u         v                        >         ?             Ĉ         ň             J         K             Ѝ         э             V         W             ܒ         ݒ              ^         c                        +         ,  j         o                        3         8  v         {                        ?         D             ť         ʥ             K         P             Ѫ         ֪             W         \             ݯ                     %  c         h                        ,         1  o         t                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               D!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
                   
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            8     -,-h     ,--     -A-    A-6-    6-CE-    CE-x-l    x-9-    9--x    -b-	    b-{-
    {--
    --|    -%ֳ-    %ֳ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               E!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         %            J  "          o  G            l                                              F	  $
          u
  S                      !         "  `         a                                 	  G         H                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |        D%  &          =  >          #?  @          O@  0A          {A  \B          B  C         I  J         J  0L         5L  sM         xM  N         N  O         O  <Q         AQ  R         R  S         S  U         
U  HV         MV  W         W  X         X  Z         Z  T[         Y[  \         \  ]         ]  _         "_  ``         e`  a         a  b         b  )d         .d  le         qe  f         f  g         g  5i         :i  xj         }j  k         k  l         m  En         Fn  o         o  p         p  
r         r  Ms         Rs  t         t  u         u  w         w  Yx         ^x  y         y  z         z  "|         '|  e}         j}  ~         ~             .         3  q         v                        :         ?  }                    ň             F         K             ̍         э             R         W             ؒ         ݒ              b         c                        '         ,  n         o                        7         8  z         {                         C         D             ɥ         ʥ             O         P             ժ         ֪             [         \                        $         %  g         h                        0         1  s         t                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               D!         T!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
                   
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            8     -,-h     ,--     -;-    ;-6-    6-yE-    yE-x-l    x-{9-    {9--x    -b-	    b-{-
    {--
    --|    -%ֳ-    %ֳ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               C!         N!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         ,            X  0            \                                              >  	          o	  J
          
  {                      N         O                                   ?         @  ~                                 ;                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |        D%  &          8  8          G9  %:          s:  Q;          ;  }<          <  =         I  J         J  3L         5L  vM         xM  N         N  O         O  ?Q         AQ  R         R  S         S  U         
U  KV         MV  W         W  X         X  Z         Z  W[         Y[  \         \  ]         ]   _         "_  c`         e`  a         a  b         b  ,d         .d  oe         qe  f         f  g         g  8i         :i  {j         }j  k         k  m         m  Bn         Fn  o         o  p         p  r         r  Ps         Rs  t         t  u         u  w         w  \x         ^x  y         y  z         z  %|         '|  h}         j}  ~         ~             1         3  t         v                        =         ?             È         ň             I         K             ύ         э             U         W             ے         ݒ              _         c                        *         ,  k         o                        4         8  w         {                        @         D             ƥ         ʥ  	           L         P             Ҫ         ֪             X         \             ޯ           !         %  d         h                        -         1  p         t                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               @!         L!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
                   
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            8     -,-h     ,--     -A-    A-6-    6-CE-    CE-x-l    x-9-    9--x    -b-	    b-{-
    {--
    --|    -%ֳ-    %ֳ--                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               E!         R!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         &            L  $          r  J            p                                              K	  *
          z
  Y                      &         '  e         f                                   L         M                        	                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |        D%  &          C  D          D  E          +F  G          WG  8H          H  dI         I  J         J  1L         5L  tM         xM  N         N  O         O  =Q         AQ  R         R  S         S  U         
U  IV         MV  W         W  X         X  Z         Z  U[         Y[  \         \  ]         ]  _         "_  a`         e`  a         a  b         b  *d         .d  me         qe  f         f  g         g  6i         :i  yj         }j  k         k  l         m  Dn         Fn  o         o  p         p  r         r  Ns         Rs  t         t  u         u  w         w  Zx         ^x  y         y  z         z  #|         '|  f}         j}  ~         ~             /         3  r         v                        ;         ?  ~                    ň             G         K             ͍         э             S         W             ْ         ݒ              a         c                        (         ,  m         o                        6         8  y         {                        B         D             ȥ         ʥ             N         P             Ԫ         ֪             Z         \                        #         %  f         h                        /         1  r         t                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               B!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M   7   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
                   
                        b                     operation.2     ipu_stack_reduce_sum_4_f16      $      E   :   E   z   I   2   I      I   Z   M   "  ]   z   a   *   a   z   e   *   e   z   i   *   i   z   m   *   m   z   q   *   q   b   u      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 4 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    input3Elements  5184    invocations     1                                    b      J      7   M   '   operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                    
         
                                b                     operation.2     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   *   Y   z   ]   *   ]   z   a   *   a   z   e   *   e   b   i      reason  OperatorKernel  value   6       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  5184    input0Elements  5184    input1Elements  5184    input2Elements  5184    invocations     1                           
         b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     
            0     -,-X     ,--     -A-    A-;-x    ;-#J-|    #J-q}-\    q}-7-    7--h    -b-    b-n-	    n--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               D!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P             >         ?  }         ~                        :                                                          4            S  +          ~  V                                                          .         /  m          n   !         !  "                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            >         @  ~                                  >         @  ~                      	          
  >         @  ~                                  >         @  ~                                  >         @  ~                                  >         @  ~                                  >         @  ~             !         !  "          #  >$         @$  ~%          =  >          >  ?          @  @          -A  B          YB  1C         K  L         L  /N         0N  oO         pO  P         P  Q         Q  /S         0S  oT         pT  U         U  V         V  /X         0X  oY         pY  Z         Z  [         [  /]         0]  o^         p^  _         _  `         `  /b         0b  oc         pc  d         d  e         e  /g         0g  oh         ph  i         i  j         j  /l         0l  om         pm  n         n  o         o  /q         0q  or           2         4  r         t                        2         4  r         t                        2         4  r         t                        2         4  r         t                        2         4  r         t                        2         4  r         t                        2         4  r         t                        2                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               G!         Z!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     +
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -M-    M-;-x    ;-J-|    J-}-\    }-#-    #--h    -b-    b-ex-	    ex--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               H!         \!  1         
2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         }                        9	         :	  x
         y
                        *         +  i         j                                               J  *          {  [                                              '              L!  $"          q"  I#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ?         @                                    ?         @                        	          
  ?         @                                    ?         @                                    ?         @                                    ?         @                                    ?         @               !         !  "          #  ?$         @$  %         %  &         &  '          (  ?)         @)  *         *  +         +  ,          -  ?.         @.  /         /  0         0  1          2  ?3         @3  4         4  5         5  6          7  ?8         @8  9         9  :         :  ;          <  ?=         @=  >         >  ?         ?  @          A  ?B         @B  C         C  D         D  E          F  ?G         @G  H         H  I         I  J          K  >L         @L  ~M         M  N         N  O          P  >Q         @Q  ~R         R  S         S  T          U  >V         @V  ~W         W  X         X  Y          Z  >[         @[  ~\         \  ]         ]  ^          _  >`         @`  ~a         a  b         b  c          d  >e         @e  ~f         f  g         g  h          i  >j         @j  ~k         k  l         l  m          Fn  8o          o  {p          p  q          r  s          Rs  Dt            ^         `                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               K!         Z!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 e
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -u -X     u -ְ-     ְ-=ܰ-    =ܰ-4-p    4-Q@-T    Q@-h-4    h-7-    7--8    -b-    b-	n-	    	n--8
    -?׳-
                   =         b      J      7   U   7   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               D!         T!  R1                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 64 })  outputElements  5184    input0Elements  31104   input1Elements  24576   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     4                           #         b      J      7   M     operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       @             >         ?  }         ~                       `	  8
          
  X            x                                              8            d  <           :         ;  y         z                                           b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     4                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  1728    input0Elements  1728    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       x            ?         @                                    ?         @                        	          
  ?         @                                    ?         @                                    ?         @                                    ?         @                                    ?         @               !         !  "          #  ?$         @$  %          C  ]D          D  E          E  F          	G  G         K  L         L  .N         0N  nO         pO  P         P  Q         Q  .S         0S  nT         pT  U         U  V         V  .X         0X  nY         pY  Z         Z  [         [  .]         0]  n^         p^  _         _  `         `  .b         0b  nc         pc  d         d  e         e  .g         0g  nh         ph  i         i  j         j  .l         0l  nm         pm  n         n  o         o  .q         0q  nr           3         4  s         t                        3         4  s         t                        3         4  s         t                        3         4  s         t                        3         4  s         t                        3         4  s         t                        3         4  s         t                        3                   >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               G!         Z!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                     *
      
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -G-    G-;-x    ;-J-|    J-}-\    }-#-    #--h    -b-    b-ex-	    ex--h
    -?׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               F!         X!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         }                        9	         :	  x
         y
                        *         +  i         j                                               J  +          {  \                                              )   !          O!  '"          u"  M#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            >         @  ~                                  >         @  ~                      	          
  >         @  ~                                  >         @  ~                                  >         @  ~                                  >         @  ~                                  >         @  ~             !         !  "          #  >$         @$  ~%         %  &         &  '          (  >)         @)  ~*         *  +         +  ,          -  >.         @.  ~/         /  0         0  1          2  >3         @3  ~4         4  5         5  6          7  >8         @8  ~9         9  :         :  ;          <  >=         @=  ~>         >  ?         ?  @          A  >B         @B  ~C         C  D         D  E          F  >G         @G  ~H         H  I         I  J          K  ?L         @L  M         M  N         N  O          P  ?Q         @Q  R         R  S         S  T          U  ?V         @V  W         W  X         X  Y          Z  ?[         @[  \         \  ]         ]  ^          _  ?`         @`  a         a  b         b  c          d  ?e         @e  f         f  g         g  h          i  ?j         @j  k         k  l         l  m          t  u          u  v          w  x          ^x  Py          y  z            _         `                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               I!         Z!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 e
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -;-    ;-6-x    6-mE-|    mE-x-\    x-y-    y--h    -b-    b-_x-	    _x--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               D!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         +            V  .            Y                                              9  	          j	  D
          
  u           
           I         J                                   :         ;  y         z                        6                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |                        <  $          x  `                                 p&  '         '  (         (  4*         9*  w+         |+  ,         ,  -         .  @/         E/  0         0  1         1  	3         3  L4         Q4  5         5  6         6  8         8  X9         ]9  :         :  ;         ;  !=         &=  d>         i>  ?         ?  @         @  -B         2B  pC         uC  D         D  E         E  9G         >G  |H         H  I         I  K         K  EL         JL  M         M  N         N  P         P  UQ         VQ  R         R  S         S  U         U  aV         bV  W         W  X         X  *Z         +Z  m[         n[  \         \  ]         ]  6_         7_  y`         z`  a         a  b          c  Bd         Cd  e         e  f         f  h         h  Ni         Oi  j         j  k         k  m         m  Zn         [n  o         o  p         p  #r         $r  fs         gs  t         t  u         u  /w         0w  rx         sx  y         y  z         z  ;|         <|  ~}         }  ~         ~             G         H             ̓         ΃             S         T             و         ڈ             _         `                        (         )  k         l                        4         5  w         x                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               F!         X!  1         2  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 c
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -G-    G-6-x    6-7E-|    7E-x-\    x-y-    y--h    -b-    b-ex-	    ex--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               H!         \!  1         
2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         %            J  "          o  G            l                                              F	  $
          u
  S                      !         "  `         a                                 	  G         H                                                      b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         X  F                                              H  6         p&  '         '  (         (  8*         9*  {+         |+  ,         ,  .         .  D/         E/  0         0  1         1  3         3  P4         Q4  5         5  6         6  8         8  \9         ]9  :         :  ;         ;  %=         &=  h>         i>  ?         ?  @         @  1B         2B  tC         uC  D         D  E         E  =G         >G  H         H  I         I  K         K  IL         JL  M         M  N         N  P         P  QQ         VQ  R         R  S         S  U         U  ]V         bV  W         W  X         X  &Z         +Z  i[         n[  \         \  ]         ]  2_         7_  u`         z`  a         a  b          c  >d         Cd  e         e  f         f  h         h  Ji         Oi  j         j  k         k  m         m  Vn         [n  o         o  p         p  r         $r  bs         gs  t         t  u         u  +w         0w  nx         sx  y         y  z         z  7|         <|  z}         }  ~         ~              C         H             Ƀ         ΃             O         T             Ո         ڈ             [         `                        $         )  g         l                        0         5  s         x                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               B!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 e
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -;-    ;-6-x    6-mE-|    mE-x-\    x-y-    y--h    -b-    b-_x-	    _x--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               D!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         ,            X  0            \                                              >  	          o	  J
          
  {                      N         O                                   ?         @  ~                                 ;                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |         ,            h  P            	          	  
                     p&  '         '  (         (  5*         9*  x+         |+  ,         ,  -         .  A/         E/  0         0  1         1  
3         3  M4         Q4  5         5  6         6  8         8  Y9         ]9  :         :  ;         ;  "=         &=  e>         i>  ?         ?  @         @  .B         2B  qC         uC  D         D  E         E  :G         >G  }H         H  I         I  K         K  FL         JL  M         M  N         N  P         P  TQ         VQ  R         R  S         S  U         U  `V         bV  W         W  X         X  )Z         +Z  l[         n[  \         \  ]         ]  5_         7_  x`         z`  a         a  b          c  Ad         Cd  e         e  f         f  
h         h  Mi         Oi  j         j  k         k  m         m  Yn         [n  o         o  p         p  "r         $r  es         gs  t         t  u         u  .w         0w  qx         sx  y         y  z         z  :|         <|  }}         }  ~         ~             F         H             ̃         ΃             R         T             ؈         ڈ             ^         `                        '         )  j         l                        3         5  v         x                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               D!         T!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 c
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -A-    A-6-x    6-=E-|    =E-x-\    x-y-    y--h    -b-    b-ex-	    ex--h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               F!         X!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                         &            L  $          r  J            p                                              K	  *
          z
  Y                      &         '  e         f                                   L         M                        	                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |           r                                  8  &          t  b         p&  '         '  (         (  7*         9*  z+         |+  ,         ,   .         .  C/         E/  0         0  1         1  3         3  O4         Q4  5         5  6         6  8         8  [9         ]9  :         :  ;         ;  $=         &=  g>         i>  ?         ?  @         @  0B         2B  sC         uC  D         D  E         E  <G         >G  H         H  I         I  K         K  HL         JL  M         M  N         N  P         P  RQ         VQ  R         R  S         S  U         U  ^V         bV  W         W  X         X  'Z         +Z  j[         n[  \         \  ]         ]  3_         7_  v`         z`  a         a  b          c  ?d         Cd  e         e  f         f  h         h  Ki         Oi  j         j  k         k  m         m  Wn         [n  o         o  p         p   r         $r  cs         gs  t         t  u         u  ,w         0w  ox         sx  y         y  z         z  8|         <|  {}         }  ~         ~             D         H             ʃ         ΃             P         T             ֈ         ڈ             \         `                        %         )  h         l                        1         5  t         x                                >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               B!         P!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                 e
        
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -A-    A-;-x    ;-YJ-|    YJ-}-\    }-7-    7--h    -cb-    cb--	    --h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               H!         \!  1         
2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 C	         D	  
         
                        *         +  i         j                                               F  *          u  Y                                                            0!  "          O"  '#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            >         C                                   J         O                        
         
  V         [                                 $  b         g                        +         0  n         s                        7         <  z                                  C         H                          "         "  O#         T#  $         $  %         %  '         '  [(         `(  )         )  *         *  $,         ),  g-         l-  .         .  /         /  01         51  s2         x2  3         3  4         4  <6         A6  7         7  8         8  :         
:  H;         M;  <         <  =         =  ?         ?  T@         Y@  A         A  B         B  D         "D  `E         eE  F         F  G         G  )I         .I  lJ         qJ  K         n  o         o  %q         &q  hr          I~  1          y  a                      ف            	           	  K         L             љ         ҙ             W         X             ݞ         ޞ            !  c         d                        ,         -  o         p                        8         9  {         |                        D         E             ʱ         ˱             P         Q             ֶ         ׶                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               7!         8!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-;-x    ;-wJ-|    wJ-}-\    }-7-    7--h    -{b-    {b--	    --h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               D!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         }                        9	         :	  x
         y
                        *         +  i         j                                               J  *          {  [                                              '              L!  $"          q"  I#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            B         C                                   N         O                        
         
  Z         [                        #         $  f         g                        /         0  r         s                        ;         <  ~                                 G         H                          "         "  S#         T#  $         $  %         %  '         '  _(         `(  )         )  *         *  (,         ),  k-         l-  .         .  /         /  41         51  w2         x2  3         3  4         4  @6         A6  7         7  8         8  	:         
:  L;         M;  <         <  =         =  ?         ?  X@         Y@  A         A  B         B  !D         "D  dE         eE  F         F  G         G  -I         .I  pJ         qJ  K         n  o         o  !q         &q  dr          )            Y  ;            k                        ˏ         	  G         L             ͙         ҙ             S         X             ٞ         ޞ           !  _         d                        (         -  k         p                        4         9  w         |                        @         E             Ʊ         ˱  	           L         Q             Ҷ         ׶                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               C!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -;-    ;-;-x    ;-YJ-|    YJ-}-\    }-7-    7--h    -ib-    ib--	    --h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               F!         X!  1         2  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P                                 C	         D	  
         
                        *         +  i         j                                               F  +          u  Z                                                            3!  "          S"  +#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            ?         C                                   K         O                        
         
  W         [                                  $  c         g                        ,         0  o         s                        8         <  {                                 D         H                          "         "  P#         T#  $         $  %         %  '         '  \(         `(  )         )  *         *  %,         ),  h-         l-  .         .  /         /  11         51  t2         x2  3         3  4         4  =6         A6  7         7  8         8  :         
:  I;         M;  <         <  =         =  ?         ?  U@         Y@  A         A  B         B  D         "D  aE         eE  F         F  G         G  *I         .I  mJ         qJ  K         n  o         o  $q         &q  gr          9  !          i  Q                      ɇ                       	  J         L             Й         ҙ             V         X             ܞ         ޞ           !  b         d                        +         -  n         p                        7         9  z         |                         C         E             ɱ         ˱             O         Q             ն         ׶                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               ;!         <!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                    0     -,-X     ,--     -5-    5-;-x    ;-}J-|    }J-}-\    }-7-    7--h    -{b-    {b--	    --h
    -9׳-
                   =         b      J      7   U   G   operation.0     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   1       invocations     1                               D!         T!  1         1  =                              b        -   w         operation.0     ipu_stack_gemm_f16_init_small_rows_interleaved_k384_c80_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   3       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 384, output_columns: 80 })  outputElements  6480    input0Elements  31104   input1Elements  30720   invocations     1                                      b            W   u      operation.0     ipu_stack_copy_u64            %   :   %   R   )   2   )      )   2   )   "   )   B   )   Z   -   b   1      reason  LocalCopy       value   3       bytes   864     pattern Contiguous      invocations     5                           #         b      J      7   M   G  operation.0     exchange                 :      z      2            b         reason  OperatorInputs  value   3       invocations     1       P         }                        9	         :	  x
         y
                        *         +  i         j                                               J  +          {  \                                              )   !          O!  '"          u"  M#                              b                     operation.0     ipu_stack_reduce_sum_3_f16             =   :   =   z   A   2   A      A   Z   E   "  U   z   Y   "   Y   z   ]   "   ]   z   a   "   a   z   e   "   e   b   i      reason  OperatorKernel  value   3       kernelSpec      Planned(ReductionSum { inputs: 3 })     outputElements  432     input0Elements  432     input1Elements  432     input2Elements  432     invocations     5                                      b            g         operation.1     ipu_stack_gelu_tanh_approx_f16        -   :   -   z   1   2   1      1   Z   5   r   9   z   =   *   =   z   A   *   A   b   E      reason  OperatorKernel  value   4       kernelSpec      Planned(Gelu)   outputElements  2160    input0Elements  2160    invocations     1                           3         b      J      7   Q     operation.2     exchange                 :            2            b   !      reason  LayoutRearrangement     value   5       invocations     1       |            A         C                        
           M         O                        
         
  Y         [                        "         $  e         g                        .         0  q         s                        :         <  }                                 F         H                          "         "  R#         T#  $         $  %         %  '         '  ^(         `(  )         )  *         *  ',         ),  j-         l-  .         .  /         /  31         51  v2         x2  3         3  4         4  ?6         A6  7         7  8         8  :         
:  K;         M;  <         <  =         =  ?         ?  W@         Y@  A         A  B         B   D         "D  cE         eE  F         F  G         G  ,I         .I  oJ         qJ  K         n  o         o  "q         &q  er                      I  +          y  [                      ٔ           	  H         L             Ι         ҙ             T         X             ڞ         ޞ           !  `         d                        )         -  l         p                        5         9  x         |                        A         E             Ǳ         ˱  
           M         Q             Ӷ         ׶                     >         b      J      7   U   G   operation.2     exchange                 :         !   2   !      !   b   %      reason  OperatorInput { input: 1 }      value   2       invocations     1                               C!         D!  1         1  =                              b        -   w         operation.2     ipu_stack_gemm_f16_init_small_rows_interleaved_k480_c64_r81_r81       5   :   5   z   9   2   9      9   Z   =     y   z   }   *   }   z      2      z      2      b         reason  OperatorKernel  value   6       kernelSpec      Planned(Gemm { multiply: F16, accumulate: F16, mode: Initialize, weights: Interleaved, inner_block: 480, output_columns: 64 })  outputElements  5184    input0Elements  38880   input1Elements  30720   invocations     1              	                      b      J      7   M      operation.2     exchange                 :      z      2            b         reason  OperatorInputs  value   6       invocations     1                         
                       b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1                                     b      *      7   I      operation.2     sync             :            2            b         reason  ExchangeBarrier value   6       invocations     1              